Topic tag determination method, terminal device, cloud server, and readable storage medium

Abstract

The present application is applicable to the technical field of terminals, and particularly relates to a topic tag determination method, a terminal device, a cloud server, and a computer-readable storage medium. In the method, after acquiring an association relationship between a first topic tag and a second topic tag both corresponding to first content, a terminal device displays the first topic tag on the basis of the association relationship; and when displaying the first topic tag, the terminal device displays, on the basis of a first operation of a user for a certain first topic tag, one or more second topic tags associated with the first topic tag. That is, a terminal device can hierarchically display topic tags corresponding to first content, so that a user can quickly select a required topic tag for the first content on the basis of the hierarchically displayed topic tags, thereby improving the efficiency of selecting the topic tags by the user, and improving the efficiency of determining the topic tags. Moreover, the user can be guided to select diverse topic tags, so that homogenous topic tags can be reduced, thereby improving the user experience.

Description

Topic label determination method, terminal device, cloud server and readable storage medium

[0001] The present application claims priority to the Chinese patent application No. 202411322350.X, filed on September 20, 2024, and entitled "Topic label determination method, terminal device, cloud server and readable storage medium", the whole content of which is incorporated herein by reference. TECHNICAL FIELD

[0002] The present application belongs to the technical field of terminals, and particularly relates to a topic label determination method, a terminal device, a cloud server and a computer readable storage medium. BACKGROUND

[0003] A topic label can be used to annotate a video or a text image, so as to summarize and describe the video or the text image, and facilitate the distribution and search of the video or the text image. For example, when publishing a video or a text image, adding a correct topic label to the video or the text image not only enables the video or the text image to be accurately distributed, but also improves the search speed and accuracy of the video or the text image.

[0004] At present, when publishing a video or a text image, recommended topic labels can be determined according to the video or the text image to be published, and all the recommended topic labels can be displayed, so that a user can select a topic label from the recommended topic labels to determine the topic label corresponding to the video or the text image. However, this way of displaying all the topic labels for the user to select has the problem of low selection efficiency, resulting in low determination efficiency of the topic label. SUMMARY

[0005] The present application provides a topic label determination method, a terminal device, a cloud server and a computer readable storage medium, which can improve the selection efficiency of the topic label and improve the determination efficiency of the topic label.

[0006] In a first aspect, the present application provides a topic label determination method applied to a terminal device, and the method comprises:

[0007] Obtaining an association relationship between a first topic label and a second topic label corresponding to first content, wherein the first topic label and the second topic label are determined according to content features corresponding to the first content;

[0008] Displaying the first topic label;

[0009] in response to a first operation, display one or more second topic labels associated with a first topic label corresponding to the first operation, the one or more second topic labels associated with the first topic label corresponding to the first operation being determined according to the association relationship;

[0010] in response to a second operation, determine one or more third topic labels corresponding to the second operation as target topic labels corresponding to the first content.

[0011] In the above-provided topic label determination method, the terminal device can obtain the association relationship between the first topic label and the second topic label corresponding to the first content, and can display the first topic label according to the association relationship. When displaying the first topic label, the terminal device can display one or more second topic labels associated with a certain first topic label based on a first operation of the user on the first topic label. That is, the terminal device can display the topic labels corresponding to the first content in a hierarchical manner, so that the user can quickly select the required topic label for the first content according to the hierarchical display of the topic labels, can reduce the number of topic labels that the user needs to view, improve the efficiency of the user selecting the topic labels, improve the efficiency of determining the topic labels, can guide the user to select a variety of topic labels, can reduce the homogenization of the topic labels, improve the accuracy of determining the topic labels, and improve the user experience.

[0012] In one example, after the display of the one or more second topic labels associated with the first topic label corresponding to the first operation, the method further includes:

[0013] in response to a third operation, display one or more fourth topic labels associated with a second topic label corresponding to the third operation, the one or more fourth topic labels associated with the second topic label corresponding to the third operation being determined according to the association relationship.

[0014] In the topic label determination method provided in this example, the topic labels corresponding to the first content can include three or more levels. When the terminal device performs hierarchical display of the topic labels corresponding to the first content, the terminal device can display the candidate topic labels of the first level (i.e., the first topic labels) by default, and can perform expanded display of the candidate topic labels of the second level or the third level (i.e., the second topic labels) based on a specified operation (e.g., the second operation or the third operation) of the user. When the user selects the topic labels corresponding to the first content, the user only needs to view the topic labels of the first level for a certain type of irrelevant topic labels, which can reduce the number of topic labels that the user needs to view during the selection of the topic labels and can improve the efficiency of the selection of the topic labels by the user. Moreover, the hierarchical display of the classified topic labels can guide the user to select a variety of topic labels and can improve the diversity and accuracy of the topic labels.

[0015] It should be noted that the association relationship between the first topic labels and the second topic labels can be represented by the hierarchical relationship between the candidate topic labels. For example, the first topic labels can include the candidate topic labels of the first level. The second topic labels can include the candidate topic labels other than the candidate topic labels of the first level.

[0016] For example, when the hierarchical relationship between the candidate topic labels includes the first level and the second level, the first topic labels can include the candidate topic labels of the first level, and the second topic labels associated with a certain first topic label can include the candidate topic labels of the second level under the first topic label. For example, when the hierarchical relationship between the candidate topic labels includes the first level, the second level, and the third level, the first topic labels can include the candidate topic labels of the first level, the second topic labels associated with a certain first topic label can include the candidate topic labels of the second level under the first topic label and the candidate topic labels of the third level under the candidate topic labels of the second level, and so on.

[0017] In some embodiments, the first operation includes a selection operation, and the third topic labels include one or more topic labels selected by the first operation.

[0018] In some embodiments, the third topic labels can include one or more of the first topic labels, and / or the third topic labels can include one or more of the second topic labels.

[0019] In the topic label determination method provided in this example, the user can select the required topic labels from the displayed first topic labels and second topic labels.

[0020] In some other embodiments, the first operation includes an input operation, and the third topic labels include one or more topic labels input by the first operation.

[0021] In the topic label determination method provided in this embodiment, the user can also customize and set the corresponding topic label for the first content, i.e., manually add the corresponding topic label for the first content. For example, when the terminal device displays the first topic label corresponding to the first content, or when the terminal device displays the first topic label and the second topic label corresponding to the first content, the terminal device can also display a button for manually adding a topic label. When the user wants to manually add a topic label, the user can click the button for manually adding a topic label to perform the customization and setting of the topic label.

[0022] In some embodiments, the obtaining of the association relationship between the first topic label and the second topic label corresponding to the first content comprises:

[0023] obtaining candidate topic labels corresponding to the first content, the candidate topic labels comprising the first topic label and the second topic label;

[0024] determining the heat and / or confidence of the candidate topic labels;

[0025] determining the association relationship between the first topic label and the second topic label according to the heat and / or confidence of the candidate topic labels.

[0026] It should be noted that the cloud server can determine relevant features similar to the content features from the first content according to the content features from the preset feature library. After determining the relevant features, the cloud server can obtain the topic labels associated with each relevant feature, and can determine the candidate topic labels corresponding to the first content according to the topic labels associated with each relevant feature. Each relevant feature can be associated with one or more topic labels. The terminal device can obtain the candidate topic labels corresponding to the first content from the cloud server. After obtaining the candidate topic labels corresponding to the first content, the terminal device can determine the first topic label and the second topic label and the association relationship between the first topic label and the second topic label according to the heat and / or confidence of the candidate topic labels.

[0027] In one example, the determining of the association relationship between the first topic label and the second topic label according to the heat and / or confidence of the candidate topic labels can comprise:

[0028] For each of the candidate topic labels, determining a score corresponding to the candidate topic label according to the heat and / or confidence of the candidate topic label;

[0029] performing descending order sorting on the candidate topic labels according to the scores to obtain a first sorting result;

[0030] According to the first ranking result, a correlation between the first topic label and the second topic label is determined.

[0031] In a possible implementation, the determining of the correlation between the first topic label and the second topic label according to the first ranking result can include:

[0032] A first candidate topic label in the first ranking result is determined as a root node, the first candidate topic label being a first candidate topic label in the first ranking result;

[0033] For a second candidate topic label in the first ranking result, a first similarity between the second candidate topic label and each root node is sequentially determined, the second candidate topic label being any one in the first ranking result and the second candidate topic label not being the first candidate topic label;

[0034] When the first similarity between the second candidate topic label and a first root node is greater than or equal to a first threshold value, the second candidate topic label is determined as a child node of the first root node, the first root node being any root node;

[0035] When the first similarity between the second candidate topic label and each root node is less than the first threshold value, the second candidate topic label is determined as a new root node;

[0036] According to each root node and a child node of each root node, a correlation between the first topic label and the second topic label is determined.

[0037] In an example, the determining of the correlation between the first topic label and the second topic label according to each root node and a child node of each root node includes:

[0038] A first candidate topic label corresponding to each root node is determined as the first topic label;

[0039] For a child node of a second root node, a second candidate topic label corresponding to the child node of the second root node is determined as a second topic label under the second root node, the second root node being any one of the root nodes.

[0040] In an example, the determining of the second candidate topic label corresponding to the child node of the second root node as the second topic label under the second root node can include:

[0041] For the child nodes of the second root node, parent-child relationships between the child nodes of the second root node are determined according to the scores corresponding to the child nodes of the second root node and second similarities between the child nodes of the second root node;

[0042] For the first parent node under the second root node, a second candidate topic label corresponding to the first parent node is determined as a second topic label under the second root node, and third candidate topic labels corresponding to the child nodes of the first parent node are determined as second topic labels under the first parent node, the first parent node being any parent node under the second root node.

[0043] In some embodiments, the acquiring of the association relationship between the first topic label and the second topic label corresponding to the first content can include:

[0044] The first content is acquired, and the first content is sent to a cloud server;

[0045] The association relationship between the first topic label and the second topic label sent by the cloud server is acquired, the first topic label and the second topic label being determined by the cloud server according to content features corresponding to the first content.

[0046] In the topic label determination method provided in this embodiment, the terminal device can determine the first topic label and the second topic label corresponding to the first content and the association relationship between the first topic label and the second topic label through the cloud server, so that the terminal device can not need to determine the topic label, and the computing overhead of the terminal device is reduced. That is, after the terminal device acquires the first content, the terminal device can directly send the first content to the cloud server. After the cloud server acquires the first content, the cloud server can determine the first topic label and the second topic label corresponding to the first content and the association relationship between the first topic label and the second topic label, and can send the determined association relationship between the first topic label and the second topic label to the terminal device.

[0047] It should be noted that the association relationship between the first topic label and the second topic label sent by the cloud server can include the first topic label and the second topic label. When the association relationship between the first topic label and the second topic label sent by the cloud server does not include the first topic label and the second topic label, the cloud server can also send the first topic label and the second topic label when sending the association relationship between the first topic label and the second topic label.

[0048] In other embodiments, the acquiring of the association relationship between the first topic label and the second topic label corresponding to the first content can include:

[0049] obtaining the first content and a first title corresponding to the first content, and sending the first content and the first title to a cloud server;

[0050] obtaining an association relationship between the first topic label and the second topic label sent by the cloud server, the first topic label and the second topic label being determined by the cloud server according to content features corresponding to the first content and title features corresponding to the first title.

[0051] In the topic label determination method provided in this embodiment, the association relationship between the first topic label and the second topic label corresponding to the first content can be determined in combination with the first title corresponding to the first content, which can improve the accuracy of the association relationship between the first topic label and the second topic label and improve user experience.

[0052] In a possible implementation, the sending the first content and the first title to the cloud server can include:

[0053] performing feature extraction on the first content to obtain content features corresponding to the first content;

[0054] performing feature extraction on the first title to obtain title features corresponding to the first title;

[0055] sending the content features and the title features to the cloud server.

[0056] In the topic label determination method provided in this implementation, in the scenario where the association relationship between the first topic label and the second topic label corresponding to the first content is determined by the cloud server, the content features corresponding to the first content and the title features corresponding to the first title can be extracted by the terminal device, so that the cloud server can directly determine the association relationship between the first topic label and the second topic label corresponding to the first content according to the content features and the title features.

[0057] In one example, when the first content is a video, the performing feature extraction on the first content to obtain content features corresponding to the first content can include:

[0058] performing frame extraction on the first content to obtain a first video frame corresponding to the first content;

[0059] performing feature extraction on the first video frame to obtain the content features corresponding to the first content.

[0060] In the topic label determination method provided in this example, when the first content is a video, the content features corresponding to the first content can include image features corresponding to part of the video frames in the first video (i.e., the first video frames obtained by frame extraction), the number of content features (i.e., image features) can be reduced, relevant features can be quickly determined, and therefore candidate topic labels corresponding to the first content can be quickly determined according to the topic labels associated with the relevant features, and the determination speed of the candidate topic labels can be improved.

[0061] In a possible implementation, the feature extraction on the first video frame to obtain the content features corresponding to the first content can include:

[0062] extracting features from the first video frame to obtain candidate features corresponding to the first content;

[0063] determining a third similarity between each candidate feature and the title feature;

[0064] sequencing the candidate features in descending order according to the third similarity to obtain a second sequencing result;

[0065] determining the content features corresponding to the first content according to the second sequencing result.

[0066] In one example, the determination of the content features corresponding to the first content according to the second sequencing result can include:

[0067] determining the first candidate feature in the second sequencing result as a key feature in a key feature set;

[0068] for each candidate feature in the second sequencing result, determining a fourth similarity between the candidate feature and each key feature in the key feature set, and determining a candidate feature with a fourth similarity smaller than a second threshold value with each key feature in the key feature set as a key feature in the key feature set;

[0069] determining the key features in the key feature set as the content features corresponding to the first content.

[0070] In the topic label determination method provided in this example, after the first video frame is obtained by frame extraction, the first video frame can be processed for duplicate removal according to the correlation between the first video frame and the first title, and irrelevant video frames such as duplicate video frames and / or transition frames can be deleted to obtain the most representative key video frame corresponding to the first content, so that the cloud server can determine the relevant features from the feature library according to the key video frame, which can not only improve the speed and efficiency of the cloud server in determining the relevant features, but also improve the accuracy of the determination of the relevant features, thereby improving the speed, efficiency and accuracy of the determination of the candidate topic labels, reducing the waiting time of the user, and improving the user experience.

[0071] In one example, the sending, to the cloud server, of the content features and the title features can include:

[0072] For each of the content features, the content feature is spliced with the title feature to obtain a spliced feature, and the spliced feature is sent to the cloud server, so that the cloud server determines the relevant features from the feature library according to each of the spliced features, and determines the first topic label and the second topic label corresponding to the first content according to the relevant topic labels associated with the relevant features.

[0073] In the topic label determination method provided in this example, after the terminal device obtains the content features and the title features, the content features and the title features can be spliced to obtain spliced features, and the spliced features can be sent to the cloud server, so that the cloud server can determine the first topic label and the second topic label corresponding to the first content according to the spliced features.

[0074] It should be understood that the splicing of the content features and the title features can also be performed by the cloud server, that is, the terminal device can send the content features and the title features to the cloud server. After the cloud server obtains the content features and the title features, the content features and the title features can be spliced to obtain spliced features.

[0075] In some embodiments, after the one or more third topic labels corresponding to the second operation are determined as the target topic labels corresponding to the first content in response to the second operation, the method can further include:

[0076] The target topic labels are sent to the cloud server, so that the cloud server updates the feature library in the cloud server according to the target topic labels.

[0077] In the topic label determination method provided in this embodiment, after the terminal device determines the target topic label corresponding to the first content, the target topic label can be sent to the cloud server. After the cloud server obtains the target topic label corresponding to the first content, the target topic label corresponding to the first content and the content features corresponding to the first content can be associated and stored in the feature library, so that the feature library can be updated according to the content features corresponding to the first content and the topic label selected or customized by the user for the first content, the accuracy of the feature library can be improved, and the accuracy of the candidate topic label can be improved.

[0078] In addition, the cloud server can also adjust the confidence of the topic label associated with the related features corresponding to the first content according to the target topic label corresponding to the first content, for example, the confidence corresponding to the topic label selected by the user can be increased, the confidence corresponding to the topic label not selected by the user can be reduced, the accuracy between the topic label and the preset feature in the feature library is improved, and the accuracy of the candidate topic label is improved.

[0079] In a second aspect, the embodiments of the present application provide a topic label determination method, applied to a cloud server, and the method comprises:

[0080] obtaining content features corresponding to first content;

[0081] determining candidate topic labels corresponding to the first content according to the content features, wherein the candidate topic labels comprise first topic labels and second topic labels;

[0082] determining an association relationship between the first topic labels and the second topic labels corresponding to the first content according to the candidate topic labels, wherein the association relationship is used to determine a target topic label corresponding to the first content.

[0083] In the above-mentioned topic label determination method, the cloud server can obtain content features corresponding to the first content, and can determine candidate topic labels corresponding to the first content according to the content features, so as to determine an association relationship between the first topic labels and the second topic labels corresponding to the first content according to the candidate topic labels.

[0084] Exemplarily, when the first content is content that the terminal device acquires and for which a topic label needs to be determined currently, after determining the association relationship between the first topic label and the second topic label corresponding to the first content, the cloud server can send the association relationship between the first topic label and the second topic label corresponding to the first content to the terminal device. The terminal device can acquire the association relationship between the first topic label and the second topic label corresponding to the first content, and can display the first topic label according to the association relationship. When the first topic label is displayed, the terminal device can display one or more second topic labels associated with the first topic label based on a first operation of a user on the first topic label. That is, the terminal device can display the topic labels corresponding to the first content in a hierarchical manner, so that the user can quickly select the required topic label for the first content according to the hierarchical display of the topic labels, the number of topic labels that the user needs to view can be reduced, the efficiency of the user selecting the topic label can be improved, the efficiency of determining the topic label can be improved, the user can be guided to select a variety of topic labels, the homogenization of the topic labels can be reduced, the accuracy of determining the topic label can be improved, and the user experience can be improved.

[0085] Exemplarily, when the first content is content without a topic label, for example, when the first content is content that has been published and does not have a topic label, after determining the association relationship between the first topic label and the second topic label corresponding to the first content, the cloud server can automatically determine the target topic label corresponding to the first content according to the association relationship between the first topic label and the second topic label corresponding to the first content, to automatically complete the topic label for the first content without a topic label.

[0086] In some embodiments, the method further comprises:

[0087] obtaining a title feature corresponding to a first title, the first title being a title of the first content;

[0088] The determining of the candidate topic label corresponding to the first content according to the content feature comprises:

[0089] The determining of the candidate topic label corresponding to the first content according to the content feature and the title feature.

[0090] In a possible implementation, the obtaining of the content feature corresponding to the first content comprises:

[0091] The first content is obtained, and feature extraction is performed on the first content to obtain the content feature corresponding to the first content;

[0092] The obtaining of the title feature of the first title comprises:

[0093] obtaining the first title and performing feature extraction on the first title to obtain title features corresponding to the first title.

[0094] In some embodiments, when the first content is a video, the performing feature extraction on the first content to obtain content features corresponding to the first content comprises:

[0095] performing frame extraction on the first content to obtain first video frames corresponding to the first content;

[0096] performing feature extraction on the first video frames to obtain content features corresponding to the first content.

[0097] In one example, the performing feature extraction on the first video frames to obtain content features corresponding to the first content comprises:

[0098] performing feature extraction on the first video frames to obtain candidate features corresponding to the first content;

[0099] determining first similarities between each of the candidate features and the title features;

[0100] performing descending order sorting on the candidate features according to the first similarities to obtain a first sorting result;

[0101] determining content features corresponding to the first content according to the first sorting result.

[0102] In one possible implementation, the determining content features corresponding to the first content according to the first sorting result comprises:

[0103] determining a first candidate feature in the first sorting result as a key feature in a key feature set;

[0104] for each candidate feature in the first sorting result, determining second similarities between the candidate feature and each key feature in the key feature set, and determining a candidate feature with second similarities to each key feature in the key feature set less than a first threshold as a key feature in the key feature set;

[0105] determining key features in the key feature set as content features corresponding to the first content.

[0106] In some embodiments, the determining candidate topic labels corresponding to the first content according to the content features and the title features comprises:

[0107] for each content feature, splicing the content feature with the title features to obtain spliced features;

[0108] For each stitching feature, a third similarity between the stitching feature and each preset feature in a feature library is determined, and according to the third similarity, a related feature corresponding to the stitching feature is determined, the feature library includes one or more preset features, and the related feature includes one or more preset features.

[0109] According to the related topic labels associated with each related feature, a candidate topic label corresponding to the first content is determined.

[0110] In some embodiments, the determining, according to the candidate topic labels, of the association relationship between the first topic label and the second topic label corresponding to the first content includes:

[0111] Obtaining the relevance between the candidate topic labels;

[0112] According to the relevance, the candidate topic labels are divided into one or more clusters, and each cluster includes one or more candidate topic labels;

[0113] For each cluster, according to the heat and / or confidence corresponding to each candidate topic label in the cluster, a score corresponding to each candidate topic label in the cluster is determined, and each candidate topic label in the cluster is arranged in descending order according to the score, to obtain a second sorting result corresponding to the cluster;

[0114] According to each second sorting result, the association relationship between the first topic label and the second topic label corresponding to the first content is determined.

[0115] In some embodiments, after the determining of the association relationship between the first topic label and the second topic label corresponding to the first content, the method further includes:

[0116] Sending the association relationship to a terminal device, so that the terminal device displays the topic label corresponding to the first content according to the association relationship after obtaining the association relationship;

[0117] Obtaining a target topic label sent by the terminal device, the target topic label including one or more topic labels selected by a user based on the topic label displayed by the terminal device, and / or including one or more topic labels set by the user;

[0118] Updating the feature library according to the target topic label, the feature library including a plurality of preset features and topic labels associated with each preset feature.

[0119] In a possible implementation, the updating of the feature library according to the target topic label includes:

[0120] For each of the content features, a target topic label corresponding to the content feature is determined according to the target topic label and a related topic label associated with a related feature corresponding to the content feature, and the content feature and the target topic label corresponding to the content feature are associated and saved to the feature library.

[0121] In another possible implementation, the updating the feature library according to the target topic label comprises:

[0122] According to the target topic label and a related topic label associated with a related feature corresponding to each of the content features, a confidence degree corresponding to the related topic label associated with the related feature is adjusted.

[0123] In some embodiments, after determining the association relationship between the first topic label and the second topic label corresponding to the first content, the method comprises:

[0124] According to the first topic label, a target topic label corresponding to the first content is determined.

[0125] In the topic label determination method provided in this embodiment, the cloud server can acquire content that has been published and does not have a topic label, and can automatically complete the target topic label for the content that does not have a topic label. For example, the cloud server can periodically perform retrieval to determine content on the network that does not have a topic label. After determining the content that does not have a topic label, the cloud server can determine candidate topic labels corresponding to the content and a hierarchical relationship between the candidate topic labels. After determining the hierarchical relationship between the candidate topic labels corresponding to the content, the cloud server can select Q candidate topic labels from the candidate topic labels of the first level (i.e., the first topic labels) as the target topic labels corresponding to the content.

[0126] In one example, the cloud server can select Q candidate topic labels from the candidate topic labels of the first level as the target topic labels corresponding to the content according to a confidence degree and / or a heat degree corresponding to each of the candidate topic labels. For example, the cloud server can determine the Q candidate topic labels of the first level with the highest confidence degrees as the target topic labels corresponding to the content. For example, the cloud server can determine the Q candidate topic labels of the first level with the highest heat degrees as the target topic labels corresponding to the content. For example, the cloud server can determine the Q candidate topic labels of the first level with the highest scores corresponding to the confidence degrees and the heat degrees as the target topic labels corresponding to the content.

[0127] In another example, the cloud server can obtain the click rate of the content corresponding to each candidate topic label of the first level, and can select Q candidate topic labels from the candidate topic labels of the first level according to the click rate as the target topic labels corresponding to the content, so as to improve the effect of automatic completion of the topic labels according to the content with good distribution effect. For example, the cloud server can determine the Q candidate topic labels of the first level corresponding to the content with the highest click rate as the target topic labels corresponding to the content.

[0128] In another example, the cloud server can obtain the publishing time of the content corresponding to each candidate topic label of the first level, and can select Q candidate topic labels from the candidate topic labels of the first level according to the publishing time of the content as the target topic labels corresponding to the content, so as to improve the probability of appearance of new topic labels and meet actual needs by automatically completing the topic labels according to the latest content. For example, the cloud server can determine the Q candidate topic labels of the first level corresponding to the content with the latest publishing time as the target topic labels corresponding to the content.

[0129] In a third aspect, the embodiments of the present application provide an interaction system, the interaction system comprising a terminal device and a cloud server;

[0130] The cloud server is configured to obtain a content feature corresponding to first content, and determine candidate topic labels corresponding to the first content according to the content feature, wherein the candidate topic labels comprise first topic labels and second topic labels.

[0131] The cloud server is further configured to determine an association relationship between the first topic labels and the second topic labels corresponding to the first content according to the candidate topic labels, and send the association relationship to the terminal device.

[0132] The terminal device is configured to obtain the association relationship, and display the first topic labels according to the association relationship.

[0133] The terminal device is further configured to display one or more second topic labels associated with the first topic labels corresponding to the first operation in response to a first operation, wherein the one or more second topic labels associated with the first topic labels corresponding to the first operation are determined according to the association relationship.

[0134] The terminal device is further configured to determine one or more third topic labels corresponding to a second operation as target topic labels corresponding to the first content in response to the second operation.

[0135] It should be noted that the terminal device can also be used to perform the related steps in the topic label determination method of any one of the first aspects described above. Similarly, the cloud server can also be used to perform the related steps in the topic label determination method of any one of the second aspects described above.

[0136] In a fourth aspect, the embodiments of the present application provide a topic label determination apparatus applied to a terminal device, the apparatus comprising:

[0137] An association relationship obtaining module, configured to obtain an association relationship between a first topic label and a second topic label corresponding to first content, the first topic label and the second topic label being determined according to content features corresponding to the first content;

[0138] A first topic label display module, configured to display the first topic label;

[0139] A second topic label display module, configured to display one or more second topic labels associated with the first topic label corresponding to a first operation in response to the first operation, the one or more second topic labels associated with the first topic label corresponding to the first operation being determined according to the association relationship;

[0140] A target topic label determination module, configured to determine one or more third topic labels corresponding to a second operation as target topic labels corresponding to the first content in response to the second operation.

[0141] In one example, the apparatus further comprises:

[0142] A fourth topic label display module, configured to display one or more fourth topic labels associated with the second topic label corresponding to a third operation in response to the third operation, the one or more fourth topic labels associated with the second topic label corresponding to the third operation being determined according to the association relationship.

[0143] In some embodiments, the first operation comprises a selection operation, and the third topic label comprises one or more topic labels selected by the first operation.

[0144] In other embodiments, the first operation comprises an input operation, and the third topic label comprises one or more topic labels input by the first operation.

[0145] In some embodiments, the association relationship obtaining module is specifically configured to obtain candidate topic labels corresponding to the first content, the candidate topic labels including the first topic label and the second topic label; determine a heat degree and / or a confidence degree corresponding to the candidate topic labels; and determine the association relationship between the first topic label and the second topic label according to the heat degree and / or the confidence degree corresponding to the candidate topic labels.

[0146] In one example, the association relationship obtaining module is further configured to, for each of the candidate topic labels, determine a score corresponding to the candidate topic label according to the heat degree and / or the confidence degree corresponding to the candidate topic label; sort the candidate topic labels in descending order according to the scores to obtain a first sorting result; and determine the association relationship between the first topic label and the second topic label according to the first sorting result.

[0147] In one possible implementation, the association relationship obtaining module is further configured to determine a first candidate topic label in the first sorting result as a root node, the first candidate topic label being a first candidate topic label in the first sorting result; for a second candidate topic label in the first sorting result, determine a first similarity between the second candidate topic label and each of the root nodes in sequence, the second candidate topic label being any one of the first sorting result and the second candidate topic label not being the first candidate topic label; when a first similarity between the second candidate topic label and a first root node is greater than or equal to a first threshold value, determine the second candidate topic label as a child node of the first root node, the first root node being any root node; when the first similarity between the second candidate topic label and each of the root nodes is less than the first threshold value, determine the second candidate topic label as a new root node; and determine the association relationship between the first topic label and the second topic label according to each of the root nodes and child nodes of each of the root nodes.

[0148] In one example, the association relationship obtaining module is further configured to determine a first candidate topic label corresponding to each of the root nodes as the first topic label; and for a child node of a second root node, determine a second candidate topic label corresponding to the child node of the second root node as a second topic label under the second root node, the second root node being any one of the root nodes.

[0149] In one example, the association relationship obtaining module is further configured to, for the child nodes of the second root node, determine the parent-child relationship between the child nodes of the second root node according to the scores corresponding to the child nodes of the second root node and the second similarities between the child nodes of the second root node; and for a first parent node under the second root node, determine the second candidate topic label corresponding to the first parent node as the second topic label under the second root node, and determine the third candidate topic label corresponding to the child nodes of the first parent node as the second topic label under the first parent node, the first parent node being any parent node under the second root node.

[0150] In some embodiments, the association relationship obtaining module is further configured to obtain the first content and send the first content to a cloud server; and obtain the association relationship between the first topic label and the second topic label sent by the cloud server, the first topic label and the second topic label being determined by the cloud server according to the content feature corresponding to the first content.

[0151] In other embodiments, the association relationship obtaining module is further configured to obtain the first content and a first title corresponding to the first content, and send the first content and the first title to a cloud server; and obtain the association relationship between the first topic label and the second topic label sent by the cloud server, the first topic label and the second topic label being determined by the cloud server according to the content feature corresponding to the first content and the title feature corresponding to the first title.

[0152] In one possible implementation, the association relationship obtaining module is further configured to perform feature extraction on the first content to obtain a content feature corresponding to the first content; perform feature extraction on the first title to obtain a title feature corresponding to the first title; and send the content feature and the title feature to the cloud server.

[0153] In one example, when the first content is a video, the association relationship obtaining module is further configured to perform frame extraction on the first content to obtain a first video frame corresponding to the first content; and perform feature extraction on the first video frame to obtain a content feature corresponding to the first content.

[0154] In one possible implementation, the association relationship obtaining module is further configured to perform feature extraction on the first video frame to obtain a candidate feature corresponding to the first content; determine a third similarity between each candidate feature and the title feature; perform descending order sorting on the candidate features according to the third similarity to obtain a second sorting result; and determine the content feature corresponding to the first content according to the second sorting result.

[0155] In an example, the association relationship obtaining module is further configured to determine a first candidate feature in the second ranking result as a key feature in a key feature set; for each candidate feature in the second ranking result, determine a fourth similarity between the candidate feature and each key feature in the key feature set, and determine a candidate feature with a fourth similarity less than a second threshold as a key feature in the key feature set; and determine the key feature in the key feature set as a content feature corresponding to the first content.

[0156] In an example, the association relationship obtaining module is further configured to, for each content feature, splice the content feature with the title feature to obtain a spliced feature, and send the spliced feature to the cloud server, so that the cloud server determines a related feature from a feature library according to each spliced feature, and determines the first topic label and the second topic label corresponding to the first content according to a related topic label associated with the related feature.

[0157] In some embodiments, the apparatus further includes:

[0158] a target topic label sending module configured to send the target topic label to the cloud server, so that the cloud server updates a feature library in the cloud server according to the target topic label.

[0159] In a fifth aspect, an embodiment of the present application provides a topic label determination apparatus applied to a cloud server, and the apparatus includes:

[0160] a content feature obtaining module configured to obtain a content feature corresponding to first content;

[0161] a topic label determination module configured to determine a candidate topic label corresponding to the first content according to the content feature, the candidate topic label including a first topic label and a second topic label;

[0162] an association relationship determination module configured to determine an association relationship between the first topic label and the second topic label corresponding to the first content according to the candidate topic label, the association relationship being used to determine a target topic label corresponding to the first content.

[0163] In some embodiments, the apparatus further includes:

[0164] a title feature obtaining module configured to obtain a title feature corresponding to a first title, the first title being a title of the first content;

[0165] the topic label determination module is further configured to determine a candidate topic label corresponding to the first content according to the content feature and the title feature.

[0166] In a possible implementation, the content feature acquisition module is further configured to acquire the first content, and perform feature extraction on the first content to obtain the content feature corresponding to the first content.

[0167] The title feature acquisition module is further configured to acquire the first title, and perform feature extraction on the first title to obtain the title feature corresponding to the first title.

[0168] In some embodiments, when the first content is a video, the content feature acquisition module is further configured to perform frame extraction on the first content to obtain a first video frame corresponding to the first content, and perform feature extraction on the first video frame to obtain the content feature corresponding to the first content.

[0169] In an example, the content feature acquisition module is further configured to perform feature extraction on the first video frame to obtain a candidate feature corresponding to the first content, determine a first similarity between each candidate feature and the title feature, sort the candidate features in descending order according to the first similarity to obtain a first sorting result, and determine the content feature corresponding to the first content according to the first sorting result.

[0170] In a possible implementation, the content feature acquisition module is further configured to determine a first candidate feature in the first sorting result as a key feature in a key feature set, determine a second similarity between each candidate feature in the first sorting result and each key feature in the key feature set, and determine a candidate feature with a second similarity smaller than a first threshold as a key feature in the key feature set, and determine the key features in the key feature set as the content feature corresponding to the first content.

[0171] In some embodiments, the topic label determination module is further configured to, for each content feature, splice the content feature and the title feature to obtain a spliced feature, determine a third similarity between each spliced feature and each preset feature in a feature library, and determine a relevant feature corresponding to the spliced feature according to the third similarity, the feature library including one or more preset features, and the relevant feature including one or more preset features, and determine a candidate topic label corresponding to the first content according to a relevant topic label associated with each relevant feature.

[0172] In some embodiments, the association relationship determining module is further configured to: obtain correlations between the candidate topic labels; divide the candidate topic labels into one or more clusters according to the correlations, each cluster including one or more of the candidate topic labels; for each cluster, determine scores of the candidate topic labels in the cluster according to the heat and / or the confidence of each candidate topic label in the cluster, and arrange the candidate topic labels in the cluster in descending order according to the scores to obtain a second ranking result corresponding to the cluster; and determine the association relationship between the first topic label and the second topic label corresponding to the first content according to the second ranking results.

[0173] In some embodiments, the apparatus further includes:

[0174] An association relationship sending module configured to send the association relationship to the terminal device, so that the terminal device displays the topic label corresponding to the first content according to the association relationship after obtaining the association relationship.

[0175] A target label obtaining module configured to obtain a target topic label sent by the terminal device, the target topic label including one or more topic labels selected by a user based on the topic label displayed by the terminal device and / or one or more topic labels set by the user.

[0176] A feature library updating module configured to update a feature library according to the target topic label, the feature library including a plurality of preset features and topic labels associated with each preset feature.

[0177] In a possible implementation, the feature library updating module is specifically configured to, for each content feature, determine a target topic label corresponding to the content feature according to the target topic label and a related topic label associated with a related feature corresponding to the content feature, and save the content feature and the target topic label corresponding to the content feature in the feature library.

[0178] In another possible implementation, the feature library updating module is further configured to adjust a confidence corresponding to a related topic label associated with each related feature according to the target topic label and the related topic label associated with the related feature corresponding to each content feature.

[0179] In some embodiments, the apparatus further includes:

[0180] A target label determining module configured to determine a target topic label corresponding to the first content according to the first topic label.

[0181] In a sixth aspect, an embodiment of the present application provides a terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and when the processor executes the computer program, the terminal device implements the topic label determination method in any of the first aspect.

[0182] In a seventh aspect, an embodiment of the present application provides a cloud server, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and when the processor executes the computer program, the cloud server implements the topic label determination method in any of the second aspect.

[0183] In an eighth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and when the computer program is executed by a computer, the computer implements the topic label determination method in any of the first aspect or the second aspect.

[0184] In a ninth aspect, an embodiment of the present application provides a computer program product, when the computer program product runs on a terminal device, the terminal device executes the topic label determination method in any of the first aspect.

[0185] In a tenth aspect, an embodiment of the present application provides a computer program product, when the computer program product runs on a cloud server, the cloud server executes the topic label determination method in any of the second aspect.

[0186] It can be understood that the beneficial effects of the third aspect to the ninth aspect can be referred to the related description of the first aspect or the second aspect, and will not be repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0187] FIG. 1 is an example diagram of displaying a topic label;

[0188] FIG. 2 is a structural schematic diagram of a terminal device provided by an embodiment of the present application;

[0189] FIG. 3 is a software architecture schematic diagram of a terminal device provided by an embodiment of the present application;

[0190] FIG. 4 is a structural schematic diagram of an interactive system to which a topic label determination method provided by an embodiment of the present application is applicable;

[0191] FIG. 5 is an application scenario schematic diagram provided by an embodiment of the present application;

[0192] FIG. 6 is an example diagram one of hierarchical relationships provided by an embodiment of the present application;

[0193] FIG. 7 is an example diagram two of a hierarchical relationship according to an embodiment of the present application;

[0194] FIG. 8 is a flow diagram of a method for determining a topic label according to an embodiment of the present application;

[0195] FIG. 9 is a flow diagram of a method for determining a topic label according to an embodiment of the present application;

[0196] FIG. 10 is a flow diagram of a method for determining a topic label according to an embodiment of the present application. DETAILED DESCRIPTION

[0197] It should be understood that the term "comprises / comprising" when used in this specification and accompanying claims, indicates the presence of stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0198] It should also be understood that the term "and / or" when used in this specification and accompanying claims, means one or the other, or both, and that the term "and / or" includes any combination of one or more of the associated listed items.

[0199] As used in this specification and any claims of the application, the terms "if" and "when" can be interpreted to mean "upon determination" or "in response to a determination" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [a described condition or event] is detected" can be interpreted to mean "upon determining" or "in response to determining" or "upon detecting [the described condition or event]" or "in response to detecting [the described condition or event]," depending on the context.

[0200] In addition, the terms "first", "second", "third", etc. in the description of the present application and the appended claims are only used to distinguish descriptions, and cannot be understood as indicating or implying relative importance.

[0201] Reference in the specification to "one embodiment" or "some embodiments" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. The appearances of the phrase "in one embodiment" or "in some embodiments" in various places in the specification are not necessarily all referring to the same embodiment, although it can. The terms "comprise", "comprising", "have", "having", "include", "including", and "contain", "containing" as used in the specification and in the claims are meant to be interpreted as "including but not limited to", unless otherwise specifically noted.

[0202] In addition, the "multiple" mentioned in the embodiments of the present application should be interpreted as two or more than two.

[0203] The steps involved in the topic label determination method provided in the embodiments of the present application are only examples, not all steps are necessarily performed, or not all contents in each information or message are optional, and can be increased or reduced as needed in use. The same step or step with the same function or message in different embodiments can be mutually referenced and learned.

[0204] The business scenarios described in the embodiments of the present application are used to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0205] The topic label can be used to label the video or text content, to summarize the video or text content, and can facilitate the distribution and search of the video or text content. For example, when publishing the video or text content, adding the correct topic label to the video or text content not only makes the video or text content can be accurately distributed, but also improves the search speed and accuracy of the video or text content when searching for the video or text content.

[0206] Among them, when publishing the video or text content, the recommended topic label can be determined according to the video or text content to be published, and all recommended topic labels can be displayed, so that the user can select the topic label from the recommended topic label to determine the topic label corresponding to the video or text content.

[0207] For example, please refer to FIG. 1, which shows an example of displaying a topic label.

[0208] When publishing content such as a video or a graphic, the terminal device can acquire the content such as a video or a graphic to be published, and can acquire a title (which can be shown as title AA in FIG. 1) corresponding to the content such as a video or a graphic to be published. After acquiring the content such as a video or a graphic to be published and the title AA, the terminal device can display the content such as a video or a graphic to be published and the title AA in a display interface, and can determine recommended topic labels according to the content such as a video or a graphic to be published. After determining the recommended topic labels, the terminal device can display the recommended topic labels in the display interface. Assuming that the recommended topic labels include "#food", "#food exploration", "#barbecue", "#skewer", "#roadside food", and "#meat sandwich", as shown in FIG. 1, the terminal device can display "#food", "#food exploration", "#barbecue", "#skewer", and a button 110 for viewing more topic labels in the display interface. The user can view more recommended topic labels through the button 110 for viewing more topic labels.

[0209] In addition, as shown in FIG. 1, the terminal device can also display a setting button 120 for a location, a setting button 130 for a public range, and a button 140 for other settings in the display interface. Among them, the user can set a published location for the content such as a video or a graphic to be published through the setting button 120 for a location, and FIG. 1 exemplarily illustrates a published location set as AAAA. The user can set a public range for the content such as a video or a graphic to be published through the setting button 130 for a public range, and FIG. 1 exemplarily illustrates a public range set as public and all people can see. The user can make other settings for the content such as a video or a graphic to be published through the button 140 for other settings.

[0210] As shown in FIG. 1, the terminal device can also display a button 150 for saving a draft and a button 160 for publishing in the display interface. The user can publish the content such as a video or a graphic through the button 160 for publishing, or can save the content such as a video or a graphic through the button 150 for saving a draft.

[0211] That is, after determining the recommended topic labels, all the recommended topic labels are generally displayed equally, so that the user can find and select the required topic labels from all the displayed topic labels. However, this equal display of all the topic labels requires the user to spend a lot of time to select the required topic labels, resulting in a low efficiency of selecting topic labels and a low efficiency of determining topic labels. In addition, this equal display of all the topic labels can easily lead to the user selecting homogenized topic labels (i.e., similar topic labels), resulting in a low accuracy of determining topic labels.

[0212] To solve the above problems, the embodiment of the present application provides a topic label determination method, a terminal device, a cloud server and a computer readable storage medium. In the method, when determining the topic label corresponding to the first content, the terminal device can obtain the association relationship between the first topic label and the second topic label corresponding to the first content, and can display the first topic label according to the association relationship. When displaying the first topic label, the terminal device can display one or more second topic labels associated with the first topic label corresponding to the first operation in response to a first operation of the user. Subsequently, the terminal device can determine one or more third topic labels corresponding to the first operation as the target topic label corresponding to the first content in response to a second operation of the user.

[0213] That is, in the embodiment of the present application, the terminal device can display the recommended topic label corresponding to the first content in a hierarchical manner according to the association relationship between the first topic label and the second topic label, so that the user can quickly select the required topic label for the first content according to the hierarchical display of the topic label, which can reduce the number of topic labels required to be viewed by the user, improve the efficiency of selecting the topic label by the user, improve the efficiency of determining the topic label, guide the user to select a variety of topic labels, reduce the homogeneity of the topic label, improve the accuracy of determining the topic label, improve the user experience, and have strong ease of use and practicality.

[0214] In the embodiment of the present application, the terminal device can be a mobile phone, a tablet computer, a wearable device, a vehicle-mounted device, a smart large screen, a notebook computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), a desktop computer, etc. The specific type of the terminal device is not limited in the embodiment of the present application.

[0215] First, the terminal device related to the embodiment of the present application is introduced. Please refer to FIG. 2, which shows a structural schematic diagram of a terminal device 200.

[0216] The terminal device 200 can include a processor 210, an external memory interface 220, an internal memory 221, a universal serial bus (USB) interface 230, a charging management module 240, a power management module 241, a battery 142, an antenna 1, an antenna 2, a mobile communication module 250, a wireless communication module 260, an audio module 270, a loudspeaker 270A, a receiver 270B, a microphone 270C, a headset interface 270D, a sensor module 280, a key 290, a camera 291, and a display screen 292, and the like. The sensor module 280 can include a pressure sensor 280A, a gyroscope sensor 280B, a barometric pressure sensor 280C, a magnetic sensor 280D, an acceleration sensor 280E, a distance sensor 280F, a proximity light sensor 280G, a fingerprint sensor 280H, a temperature sensor 280J, a touch sensor 280K, an ambient light sensor 280L, a bone conduction sensor 280M, and the like.

[0217] It can be understood that the structure shown in the embodiments of the present application does not constitute a specific limitation on the terminal device 200. In other embodiments of the present application, the terminal device 200 can include more or fewer components than shown, or combine certain components, or split certain components, or different component arrangements. The components shown can be implemented in hardware, software, or a combination of software and hardware.

[0218] The processor 210 can include one or more processing units, for example: the processor 210 can include an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural-network processing unit (NPU), and the like. Different processing units can be independent devices, or can be integrated into one or more processors.

[0219] The controller can generate operation control signals according to instruction operation codes and timing signals, and complete the control of fetching instructions and executing instructions.

[0220] The processor 210 can also be provided with a memory for storing instructions and data. In some embodiments, the memory in the processor 210 is a cache memory. The memory can hold instructions or data that the processor 210 has just used or is using in a loop. If the processor 210 needs to use the instructions or data again, it can be directly called from the memory. This avoids repeated access and reduces the waiting time of the processor 210, thereby improving the efficiency of the system.

[0221] In some embodiments, the processor 210 can include one or more interfaces. The interfaces can include an inter-integrated circuit (I2C) interface, an inter-integrated circuit sound (I2S) interface, a pulse code modulation (PCM) interface, a universal asynchronous receiver / transmitter (UART) interface, a mobile industry processor interface (MIPI), a general-purpose input / output (GPIO) interface, a subscriber identity module (SIM) interface, and / or a universal serial bus (USB) interface, etc.

[0222] The USB interface 230 is an interface that conforms to the USB standard specification, and can be a Mini USB interface, a Micro USB interface, a USB Type C interface, etc. The USB interface 230 can be used to connect a charger to charge the terminal device 200, and can also be used to transmit data between the terminal device 200 and a peripheral device. It can also be used to connect a headset to play audio through the headset. The interface can also be used to connect other terminal devices, such as AR devices, etc.

[0223] It can be understood that the interface connection relationship between the modules shown in the embodiments of the present application is only illustrative and does not constitute a structural limitation of the terminal device 200. In some other embodiments of the present application, the terminal device 200 can also use different interface connection methods or combinations of multiple interface connection methods in the above embodiments.

[0224] The charging management module 240 is configured to receive a charging input from a charger.

[0225] The power management module 241 is configured to connect the battery 142 and the charging management module 240 to the processor 210. The power management module 241 receives input from the battery 142 and / or the charging management module 240 to power the processor 210, the internal memory 221, the display screen 292, the camera 291, and the wireless communication module 260.

[0226] The wireless communication function of the terminal device 200 can be implemented by the antenna 1, the antenna 2, the mobile communication module 250, the wireless communication module 260, the modem processor, and the baseband processor.

[0227] The antenna 1 and the antenna 2 are configured to transmit and receive electromagnetic wave signals. Each antenna in the terminal device 200 can be configured to cover a single or multiple communication frequency bands. Different antennas can also be multiplexed to improve the utilization of the antennas. For example, the antenna 1 can be multiplexed as a diversity antenna for a wireless local area network. In some other embodiments, the antennas can be used in combination with a tuning switch.

[0228] The mobile communication module 250 can provide a solution for wireless communication including 2G / 3G / 4G / 5G, etc. applied to the terminal device 200. The mobile communication module 250 can include at least one filter, a switch, a power amplifier, a low noise amplifier (LNA), etc. The mobile communication module 250 can receive electromagnetic waves from the antenna 1, filter, amplify, etc. the received electromagnetic waves, and transmit the processed signals to the modem processor for demodulation. The mobile communication module 250 can also amplify the signals modulated by the modem processor and convert the signals into electromagnetic waves for radiation through the antenna 1. In some embodiments, at least part of the functional modules of the mobile communication module 250 can be arranged in the processor 210. In some embodiments, at least part of the functional modules of the mobile communication module 250 and at least part of the modules of the processor 210 can be arranged in the same device.

[0229] The modem processor can include a modulator and a demodulator. The modulator is configured to modulate a low-frequency baseband signal to be transmitted into a medium-high frequency signal. The demodulator is configured to demodulate a received electromagnetic wave signal into a low-frequency baseband signal. The demodulator then transmits the demodulated low-frequency baseband signal to the baseband processor for processing. The low-frequency baseband signal processed by the baseband processor is transmitted to the application processor. The application processor outputs a sound signal through an audio device (not limited to the loudspeaker 270A and the microphone 270B, etc.) or displays an image or a video through the display screen 292. In some embodiments, the modem processor can be an independent device. In some other embodiments, the modem processor can be independent of the processor 210 and arranged in the same device as the mobile communication module 250 or other functional modules.

[0230] The wireless communication module 260 can provide a solution for wireless communication, including wireless local area networks (WLAN) (e.g., wireless fidelity (Wi-Fi) network), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), infrared (IR) technology, etc., which are applied on the terminal device 200. The wireless communication module 260 can be one or more devices integrated with at least one communication processing module. The wireless communication module 260 receives electromagnetic waves via the antenna 2, frequency-modulates and filters the electromagnetic wave signals, and transmits the processed signals to the processor 210. The wireless communication module 260 can also receive signals to be transmitted from the processor 210, frequency-modulate them, amplify them, and radiate them as electromagnetic waves via the antenna 2.

[0231] In some embodiments, antenna 1 of terminal device 200 is coupled with mobile communication module 250, and antenna 2 is coupled with wireless communication module 260, so that terminal device 200 can communicate with a network and other devices through wireless communication technologies. The wireless communication technologies can include global system for mobile communications (GSM), general packet radio service (GPRS), code division multiple access (CDMA), wideband code division multiple access (WCDMA), time-division code division multiple access (TD-SCDMA), long term evolution (LTE), BT, GNSS, WLAN, NFC, FM, and / or IR technologies, etc. The GNSS can include global positioning system (GPS), global navigation satellite system (GLONASS), beidou navigation satellite system (BDS), quasi-zenith satellite system (QZSS), and / or satellite based augmentation systems (SBAS).

[0232] Terminal device 200 implements display functions through a GPU, display screen 292, and an application processor, etc. The GPU is a microprocessor for image processing, connected with the display screen 292 and the application processor. The GPU is used to perform mathematical and geometric calculations for graphics rendering. Processor 210 can include one or more GPUs that execute program instructions to generate or change display information.

[0233] The display screen 292 is configured to display images, videos, and the like. The display screen 292 includes a display panel. The display panel can be a liquid crystal display (LCD), an organic light-emitting diode (OLED), an active-matrix organic light-emitting diode (AMOLED), a flex light-emitting diode (FLED), a Miniled, a MicroLed, a Micro-oLed, a quantum dot light emitting diodes (QLED), or the like. In some embodiments, the terminal device 200 can include one or N display screens 292, where N is a positive integer greater than 1.

[0234] The terminal device 200 can implement the photographing function through the ISP, the camera 291, the video codec, the GPU, the display screen 292, and the application processor, and the like.

[0235] The ISP is configured to process data fed back by the camera 291.

[0236] The camera 291 is configured to capture still images or videos. In some embodiments, the terminal device 200 can include one or N cameras 291, where N is a positive integer greater than 1.

[0237] The digital signal processor is configured to process digital signals, in addition to processing digital image signals, other digital signals can also be processed. For example, when the terminal device 200 selects a frequency point, the digital signal processor is configured to perform Fourier transform on the frequency point energy, and the like.

[0238] The video codec is configured to compress or decompress digital videos. The terminal device 200 can support one or more video codecs. In this way, the terminal device 200 can play or record videos in multiple encoding formats, such as moving picture experts group (MPEG) 1, MPEG 2, MPEG 3, MPEG 4, and the like.

[0239] The NPU is a neural-network (NN) calculation processor. By drawing on the structure of a biological neural network, for example, by drawing on the transmission mode between human brain neurons, the NPU can quickly process input information and can also constantly self-learn. Through the NPU, intelligent cognition and other applications of the terminal device 200 can be implemented, such as image recognition, face recognition, voice recognition, text understanding, and the like.

[0240] The external memory interface 220 can be configured to connect an external memory card, such as a Micro SD card, to extend the storage capability of the terminal device 200. The external memory card communicates with the processor 210 through the external memory interface 220 to implement a data storage function. For example, files such as music and videos are stored in the external memory card.

[0241] The internal memory 221 can be configured to store computer executable program codes including instructions. The internal memory 221 can include a program storage area and a data storage area. The program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), and the like. The data storage area can store data (such as audio data, a phonebook, etc.) created during use of the terminal device 200, and the like. In addition, the internal memory 221 can include a high-speed random access memory, and can further include a non-volatile memory such as at least one magnetic disk storage device, a flash memory device, a universal flash storage (UFS), and the like. The processor 210 executes various function applications and data processing of the terminal device 200 by running instructions stored in the internal memory 221 and / or instructions stored in a memory disposed in the processor.

[0242] The terminal device 200 can implement an audio function through the audio module 270, the speaker 270A, the receiver 270B, the microphone 270C, the earphone interface 270D, the application processor, and the like. For example, music playing, recording, and the like.

[0243] The audio module 270 is configured to convert digital audio information into an analog audio signal output, and is also configured to convert an analog audio input into a digital audio signal. The audio module 270 can also be configured to encode and decode an audio signal. In some embodiments, the audio module 270 can be disposed in the processor 210, or part of the function modules of the audio module 270 can be disposed in the processor 210.

[0244] The software system of the terminal device 200 can adopt a layered architecture, an event-driven architecture, a micro-kernel architecture, a micro-service architecture, or a cloud architecture. For example, the software system of the terminal device 200 can adopt an Android operating system (OS), a Harmony OS, an IOS, or the like with a layered architecture. Embodiments of the present application exemplarily illustrate the software structure of the terminal device 200 with a layered architecture.

[0245] FIG. 3 is a software structure block diagram of the terminal device 200 according to an embodiment of the present application.

[0246] The layered architecture divides software into several layers, each of which has a clear role and division of labor. Layers communicate with each other through software interfaces. In some embodiments, the operating system is divided into four layers, from top to bottom, the application layer, the application framework layer, the runtime and system library, and the kernel layer.

[0247] The application layer can include a series of application packages.

[0248] As shown in FIG. 3, the application packages can include camera, gallery, calendar, call, map, navigation, WLAN, Bluetooth, music, video, short message, etc.

[0249] The application framework layer provides application programming interfaces (APIs) and programming frameworks for the applications of the application layer. The application framework layer includes some pre-defined functions.

[0250] As shown in FIG. 3, the application framework layer can include window manager, content provider, view system, phone manager, resource manager, notification manager, etc.

[0251] The window manager is used to manage window programs. The window manager can obtain the size of the display screen, determine whether there is a status bar, lock the screen, and take screenshots, etc.

[0252] The content provider is used to store and obtain data, and make the data accessible to the application. The data can include video, image, audio, dialed and received calls, browsing history and bookmarks, phonebook, etc.

[0253] The view system includes visual controls, such as controls for displaying text, controls for displaying pictures, etc. The view system can be used to build applications. A display interface can be composed of one or more views. For example, a display interface including a short message notification icon can include a view for displaying text and a view for displaying pictures.

[0254] The phone manager is used to provide the communication function of the terminal device 200. For example, the management of call state (including call connection, call hang-up, etc.).

[0255] The resource manager provides various resources for the application, such as localized strings, icons, pictures, layout files, video files, etc.

[0256] The notification manager enables applications to display notification information in the status bar, which can be used to convey a message of the notification type, which can automatically disappear after a short stay without user interaction. For example, the notification manager is used to notify the completion of the download, message reminders, etc. The notification manager can also be a notification that appears in the form of a chart or a scroll bar text in the top status bar of the system, such as a notification of an application running in the background, and can also be a notification that appears in the form of a dialog window on the screen. For example, the status bar prompts text information, issues a prompt sound, the terminal device vibrates, the indicator light flashes, etc.

[0257] The runtime includes a core library and a virtual machine. The runtime is responsible for scheduling and management of the operating system.

[0258] The core library includes two parts: one part is the function function that the java language needs to call, and the other part is the core library of the operating system.

[0259] The application layer and the application framework layer run in the virtual machine. The virtual machine executes the java file of the application layer and the application framework layer into a binary file. The virtual machine is used to perform the management of the object life cycle, the management of the stack, the management of the thread, the management of the security and the exception, and the garbage collection, etc.

[0260] The system library can include multiple functional modules. For example: a surface manager, media libraries, a three-dimensional graphics processing library (for example: OpenGL ES), a 2D graphics engine (for example: SGL), etc.

[0261] The surface manager is used to manage the display subsystem, and provides a fusion of 2D and 3D layers for multiple applications.

[0262] The media library supports multiple commonly used audio, video format playback and recording, and static image files, etc. The media library can support multiple audio and video coding formats, such as: MPEG4, H.264, MP3, AAC, AMR, JPG, PNG, etc.

[0263] The three-dimensional graphics processing library is used to realize three-dimensional graphics drawing, image rendering, synthesis, and layer processing, etc.

[0264] The 2D graphics engine is a drawing engine for 2D drawing.

[0265] The kernel layer is a layer between hardware and software. The kernel layer at least includes a display driver, a camera driver, an audio driver, and a sensor driver.

[0266] It should be noted that the scheme provided by the embodiments of the present application is used to determine the topic label corresponding to the content, which is only used for exemplary explanation, and should not be understood as a limitation on the embodiments of the present application. It should be understood that the scheme provided by the embodiments of the present application can also be used in the scene of determining the title of the content, or in the scene of classifying the content, etc. For example, when determining the title of a certain content, the hierarchical relationship between the candidate titles corresponding to the content can be determined based on the scheme provided by the embodiments of the present application, and the display of the candidate titles can be performed according to the hierarchical relationship, so that the user can determine the title of the content according to the hierarchical display of the candidate titles. For example, when classifying the content, the hierarchical relationship between the contents can be determined based on the scheme provided by the embodiments of the present application, so as to classify the content according to the hierarchical relationship. Hereinafter, the topic label will be exemplarily explained.

[0267] The topic label determination method provided by the embodiments of the present application will be described in detail below in combination with the drawings and specific application scenarios.

[0268] Please refer to FIG. 4, which shows a structural schematic diagram of an interactive system to which the topic label determination method provided by the embodiments of the present application is applicable.

[0269] As shown in FIG. 4, the interactive system can include a terminal device 410 and a cloud server 420. The terminal device 410 can be in communication connection with the cloud server 420 through a communication network. The terminal device 410 can be a mobile phone, a smart television, a tablet computer, a notebook computer, or a smart watch, etc. terminal device with a display screen.

[0270] In some embodiments, when publishing the content, the terminal device 410 can obtain the content to be published (hereinafter referred to as the first content) and the title of the first content (hereinafter referred to as the first title), and can send the first content and the first title to the cloud server 420. It should be understood that the first content can be the content selected or input by the user in the content publishing process, for example, it can be a video or a text image (i.e. a content containing an image and a text) selected by the user, or it can be a text input by the user. The first title can be the title of the first content itself, or it can be the title added by the user for the first content in the content publishing process, for example, it can be the title edited by the user for the first content in the content publishing process.

[0271] After obtaining the first content and the first title, the cloud server 420 can determine the candidate topic label corresponding to the first content and the hierarchical relationship between the candidate topic labels according to the first content and the first title, and can send the candidate topic label and the hierarchical relationship between the candidate topic labels to the terminal device 410.

[0272] Alternatively, after the cloud server 420 acquires the first content and the first title, the cloud server 420 can determine the candidate topic labels corresponding to the first content according to the first content and the first title, and can send the candidate topic labels corresponding to the first content to the terminal device 410. After the terminal device 410 acquires the candidate topic labels, the terminal device 410 can determine the hierarchical relationship between the candidate topic labels.

[0273] After the terminal device 410 acquires the candidate topic labels and the hierarchical relationship between the candidate topic labels, the terminal device 410 can display the candidate topic labels hierarchically according to the hierarchical relationship between the candidate topic labels, so that the user can quickly select the required topic label according to the hierarchically displayed candidate topic labels, and the efficiency of determining the topic label is improved.

[0274] It should be noted that the hierarchical relationship between the candidate topic labels can be used to indicate the grouping relationship between the candidate topic labels and the subordinate relationship between the candidate topic labels in each group. The hierarchical relationship between the candidate topic labels can include at least two levels, and each level can include one or more candidate topic labels. That is, the candidate topic labels can be divided into one or more clusters, each cluster can include one or more candidate topic labels, and the candidate topic labels in each cluster can be divided into one or more levels.

[0275] It should be understood that the above-mentioned cloud server determining the candidate topic labels corresponding to the first content according to the first content and the first title is only an exemplary explanation, and should not be understood as a limitation of the embodiments of the present application. In the embodiments of the present application, the cloud server 420 can also determine the candidate topic labels corresponding to the first content only according to the first content.

[0276] In one example, when the hierarchical relationship between the candidate topic labels includes two levels, for example, when including a first level and a second level, when displaying the candidate topic labels according to the hierarchical relationship between the candidate topic labels, the terminal device 410 can display the candidate topic labels of the first level and the selection buttons corresponding to each candidate topic label by default. When detecting a designation operation on a certain candidate topic label (for example, candidate topic label A) of the first level, the terminal device 410 can display the candidate topic labels of the second level under the candidate topic label A and the corresponding selection buttons.

[0277] For example, when the hierarchical relationship between the candidate topic labels includes three or more levels, when the second-level candidate topic labels and the corresponding selection buttons are displayed, the terminal device 410 can display the third-level candidate topic labels under a certain candidate topic label (e.g., candidate topic label B) of the second level and the corresponding selection buttons when detecting a designation operation on the candidate topic label B. Similarly, when the third-level candidate topic labels and the corresponding selection buttons are displayed, the terminal device 410 can display the fourth-level candidate topic labels under a certain candidate topic label (e.g., candidate topic label C) of the third level and the corresponding selection buttons when detecting a designation operation on the candidate topic label C, and so on.

[0278] That is, when performing hierarchical display of the candidate topic labels, the terminal device 410 can display the first-level candidate topic labels by default, and can perform expanded display of the second-level or third-level candidate topic labels based on the designation operation of the user, so that the user only needs to view the first-level candidate topic labels when selecting the corresponding topic label for the first content for a certain category of irrelevant candidate topic labels, the number of topic labels that need to be viewed by the user during the selection of the topic label can be reduced, and the efficiency of the user in selecting the topic label can be improved. Moreover, through hierarchical display of the classified topic labels, the user can be guided to select a variety of topic labels, and the diversity and accuracy of the topic labels can be improved.

[0279] It should be noted that the designation operation can be determined according to the actual application scenario, and the embodiments of the present application do not limit this. For example, the designation operation can be a click operation, a touch operation, or a long press operation, etc. according to the actual application scenario. It should be understood that the long press operation can refer to an operation in which the touch time is greater than or equal to a preset time threshold. The preset time threshold can be determined according to the actual application scenario, for example, the preset time threshold can be any value such as 2 seconds, 3 seconds, or 4 seconds, etc. according to the actual application scenario. Hereinafter, the designation operation will be exemplarily described as a click operation.

[0280] In a possible implementation, the user can also customize the corresponding topic label for the first content, that is, manually add the corresponding topic label for the first content. For example, the terminal device can display a button for manually adding a topic label when displaying the candidate topic labels corresponding to the first content. When the user wants to manually add a topic label, the user can click the button for manually adding a topic label to perform customization of the topic label.

[0281] Please refer to FIG. 5, which shows an application scenario one provided by the embodiment of the present application. The application scenario is determined by the cloud server 420 based on the hierarchical relationship between the candidate topic labels, and the hierarchical relationship between the candidate topic labels includes a first level and a second level, which are exemplarily illustrated. In the application scenario, the first content can be the content selected by the user, and the first title can be the title added by the user when publishing the content.

[0282] As shown in (a) of FIG. 5, when publishing the content, the user can open the publishing interface. The publishing interface can include a button 510 of selecting works, a button 520 of saving drafts, a button 530 of publishing, and an input box 540 of inputting titles. The user can click the button 510 of selecting works to select the first content to be published, and can add a corresponding title to the first content through the input box 540. It is assumed that the first content selected by the user is a video AA, and the added title is a title BB.

[0283] It should be understood that the publishing interface can also include a setting button of the location, a setting button of the public range, and a button of other settings. The user can add the location of publishing to the first content through the setting button of the location, and (a) of FIG. 5 exemplarily illustrates the added location as BBBB. The user can set the public range of the first content through the setting button of the public range, and (a) of FIG. 5 exemplarily illustrates the public range as public, visible to all. The user can set other settings for the first content through the button of other settings.

[0284] As shown in (b) of FIG. 5, after obtaining the video AA selected by the user, the terminal device 410 can display the related interface of the video AA in the publishing interface, for example, can display the first frame of the video AA or display the cover of the video AA, etc. In addition, after obtaining the title BB added by the user to the video AA, the terminal device 410 can also display the title BB in the publishing interface. In addition, the terminal device 410 can send the video AA and the title BB to the cloud server 420.

[0285] After obtaining the video AA and the title BB, the cloud server 420 can determine the candidate topic label corresponding to the video AA and the hierarchical relationship between the candidate topic labels based on the video AA and the title BB, and can send the candidate topic label and the hierarchical relationship between the candidate topic labels to the terminal device 410.

[0286] It is assumed that the cloud server 420 determines, according to the video AA and the title BB, that the candidate topic labels corresponding to the video AA include "#food", "#net red", "#smoke and fire in the world", "#food store hunting", "#barbecue", "#fried skewers", "#roadside stall food", "#live broadcast", and "#goods carrying", and that the hierarchical relationship between the candidate topic labels is that "#food", "#net red", and "#smoke and fire in the world" are first-level candidate topic labels, "#food store hunting", "#barbecue", "#fried skewers", and "#roadside stall food" are second-level candidate topic labels under "#food", and "#live broadcast" and "#goods carrying" are second-level candidate topic labels under "#net red".

[0287] As shown in (b) of FIG. 5, after the terminal device 410 acquires the candidate topic labels corresponding to the video AA and the hierarchical relationship between the candidate topic labels, the terminal device 410 can display the candidate topic labels corresponding to the video AA in a hierarchical manner according to the hierarchical relationship between the candidate topic labels, that is, can display "#food", "#net red", and "#smoke and fire in the world" which are first-level candidate topic labels in the publishing interface, and can display the selection buttons corresponding to "#food", "#net red", and "#smoke and fire in the world". When the user wants to select a certain candidate topic label for the video AA, the user can click the selection button corresponding to the candidate topic label to select the candidate topic label. (b) of FIG. 5 and (c) of FIG. 5 can represent the selection buttons by rectangular boxes. That is, the user can select a certain candidate topic label by clicking the rectangular box.

[0288] When the user clicks a certain candidate topic label of the first level, as shown in (b) of FIG. 5, when the user clicks "#food", the terminal device 410 can display the second-level candidate topic labels under "#food" and the selection buttons corresponding to the candidate topic labels in the publishing interface. That is, as shown in (c) of FIG. 5, the terminal device 410 can display "#food store hunting", "#barbecue", "#fried skewers", and "#roadside stall food" in the publishing interface, and can display the selection buttons corresponding to "#food store hunting", "#barbecue", "#fried skewers", and "#roadside stall food".

[0289] As shown in (b) of FIG. 5 and (c) of FIG. 5, when the terminal device 410 displays the candidate topic labels in a hierarchical manner in the publishing interface according to the hierarchical relationship between the candidate topic labels corresponding to the video AA, the terminal device 410 can also display the button 550 for manually adding a topic label in the publishing interface. As shown in (c) of FIG. 5, when the user wants to manually add a topic label, the user can click the button 550 for manually adding a topic label to customize the topic label.

[0290] As shown in (d) of FIG. 5, after detecting the click operation on the button 550 for manually adding the topic label, the terminal device 410 can display a pop-up window to manually add the topic label through the pop-up window, i.e., the user can input the topic label to be added through the pop-up window. The pop-up window can include an input box 561, a confirm button 562, and a cancel button 563. The user can input the topic label to be added through the input box 561. After completing the input of the topic label to be added, the user can click the confirm button 562 to complete the manual addition of the topic label.

[0291] It should be understood that the manual addition of the topic label through the pop-up window as shown in (d) of FIG. 5 is only illustrative and should not be construed as a limitation on the embodiments of the present application. In the embodiments of the present application, after detecting the click operation on the button 550 for manually adding the topic label, the terminal device 410 can also jump to a topic label adding interface. The topic label adding interface can include an input box, a confirm button, and a cancel button. The user can input the topic label to be added through the input box of the topic label adding interface. After completing the input of the topic label to be added, the user can click the confirm button of the topic label adding interface to complete the manual addition of the topic label.

[0292] It should be understood that the manual addition of the topic label through the pop-up window as shown in (d) of FIG. 5 is only illustrative and should not be construed as a limitation on the embodiments of the present application. In the embodiments of the present application, after detecting the click operation on the button 550 for manually adding the topic label, the terminal device 410 can also jump to a topic label adding interface. The topic label adding interface can include an input box, a confirm button, and a cancel button. The user can input the topic label to be added through the input box of the topic label adding interface. After completing the input of the topic label to be added, the user can click the confirm button of the topic label adding interface to complete the manual addition of the topic label.

[0293] The process of determining the candidate topic label corresponding to the first content by the cloud server 420 according to the first content and the first title will be described in detail below.

[0294] In a possible implementation, the cloud server 420 can be provided with a feature library. The feature library can include a plurality of preset features and topic labels associated with each preset feature. After obtaining the first content and the first title, the cloud server 420 can extract features from the first content to obtain content features, and extract features from the first title to obtain title features. Then, the cloud server 420 can determine a preset feature (which can be referred to as a relevant feature below) similar to the first content from the feature library according to the content features corresponding to the first content and the title features corresponding to the first title, and determine the candidate topic label corresponding to the first content according to the topic label associated with the relevant feature.

[0295] It should be understood that the cloud server 420 can also determine the candidate topic label corresponding to the first content according to the first content only. That is, the cloud server 420 can also determine the preset feature similar to the first content (i.e., the relevant feature) from the feature library according to the content feature corresponding to the first content, and can determine the candidate topic label corresponding to the first content according to the topic label associated with the relevant feature.

[0296] It should be noted that the feature library including the plurality of preset features and the topic label associated with each preset feature is only illustrative and should not be construed as limiting the embodiments of the present application. In the embodiments of the present application, the association between each preset feature in the feature library and the topic label can also be performed in other manners. For example, the feature library can include a plurality of preset features and the identity (identity, ID) of the preset content associated with each preset feature, and the preset content can be associated with the topic label, that is, the cloud server 420 can determine the topic label associated with the preset feature according to the ID of the preset content associated with the preset feature.

[0297] It should be understood that the embodiments of the present application do not limit the dimensions of the content feature, the title feature and the preset feature, and the dimensions can be determined according to the actual application scenario. For example, the dimensions of the content feature, the dimensions of the title feature and the dimensions of the preset feature can all be 128 according to the actual application scenario. That is, the content feature can be represented by a 1*128 floating point vector, the title feature can also be represented by a 1*128 floating point vector, and the preset feature can also be represented by a 1*128 floating point vector.

[0298] In a possible implementation, the content feature corresponding to the first content can include one or more. Wherein, for each content feature, the cloud server 420 can splice the content feature and the title feature to obtain a spliced feature, and can determine the relevant feature corresponding to the first content from the feature library according to the spliced feature, so as to determine the candidate topic label corresponding to the first content according to the topic label associated with the relevant feature.

[0299] In an example, when the first content is a graphic text, the content feature corresponding to the first content can include an image feature and a text feature.

[0300] Exemplarily, after obtaining the first content and the first title, the cloud server 420 can perform feature extraction on each image in the first content respectively to obtain image features corresponding to the images, and can perform feature extraction on the text in the first content to obtain one or more text features. In addition, the cloud server 420 can perform feature extraction on the first title to obtain a title feature corresponding to the first title. Subsequently, for each image feature, the cloud server 420 can splice the image feature with the title feature to obtain a spliced feature. Similarly, for each text feature, the cloud server 420 can splice the text feature with the title feature to obtain a spliced feature. For each spliced feature, the cloud server 420 can determine a similarity between the spliced feature and each preset feature in the feature library, and determine a relevant feature corresponding to the spliced feature according to the similarity, so as to determine the candidate topic label corresponding to the first content according to the relevant features corresponding to the spliced features.

[0301] In another example, when the first content is text, the content feature corresponding to the first content can include a text feature.

[0302] Exemplarily, after obtaining the first content and the first title, the cloud server 420 can perform feature extraction on the first content to obtain one or more text features. In addition, the cloud server 420 can perform feature extraction on the first title to obtain a title feature. For each text feature, the cloud server 420 can splice the text feature with the title feature to obtain a spliced feature, and can determine a similarity between the spliced feature and each preset feature in the feature library to determine a relevant feature corresponding to the spliced feature according to the similarity, so as to determine the candidate topic label corresponding to the first content according to the relevant features corresponding to the spliced features.

[0303] In another example, when the first content is video, the content feature corresponding to the first content can include an image feature.

[0304] Exemplarily, after obtaining the first content and the first title, the cloud server 420 can perform feature extraction on the first content (i.e., video frames corresponding to the first content) to obtain image features corresponding to the video frames. In addition, the cloud server 420 can perform feature extraction on the first title to obtain a title feature. For each image feature, the cloud server 420 can splice the image feature with the title feature to obtain a spliced feature, and can determine a similarity between the spliced feature and each preset feature in the feature library to determine a relevant feature corresponding to the spliced feature according to the similarity, so as to determine the candidate topic label corresponding to the first content according to the relevant features corresponding to the spliced features.

[0305] It should be understood that the cloud server 420 can include one or more feature libraries. When the cloud server 420 includes one feature library, the feature library can include different types of features (such as image features and text features, etc.), that is, in different scenarios (that is, the first content is a graphic text, text or video, etc.), the cloud server 420 can determine the relevant features according to the feature library. When the cloud server 420 includes multiple feature libraries, for example, a feature library A suitable for a graphic text scenario, a feature library B suitable for a text scenario, and a feature library C suitable for a video scenario, etc., when determining the relevant features, the cloud server 420 can determine the relevant features from the corresponding feature library according to the type of the first content. For example, when the first content is a graphic text, the cloud server 420 can determine the relevant features according to the feature library A. When the first content is text, the cloud server 420 can determine the relevant features according to the feature library B. For example, when the first content is a video, the cloud server 420 can determine the relevant features according to the feature library C, etc.

[0306] It should be noted that the present application embodiment does not limit the specific way of determining the similarity between two features. It can be determined according to the actual application scenario. For example, the similarity between two features can be determined by calculating the distance (such as cosine distance) between the two features. That is, for each splicing feature, the cloud server 420 can calculate the distance (such as cosine distance) between the splicing feature and each preset feature in the feature library, and can determine the similarity between the splicing feature and each preset feature in the feature library according to the cosine distance. The value range of the cosine distance can be [0, 1]. The similarity can be (1-cosine distance). It should be understood that the smaller the cosine distance, the greater the similarity between the two features, that is, the more similar the two features; the greater the cosine distance, the smaller the similarity between the two features, that is, the less similar the two features.

[0307] The above-mentioned feature extraction of the first content, feature extraction of the first title, and splicing of the content features and the title features are performed by the cloud server 420, which is only exemplary explanation and should not be understood as a limitation of the present application embodiment. In the present application embodiment, the terminal device 410 can also perform feature extraction of the first content and feature extraction of the first title, and the cloud server 420 can perform splicing of the content features and the title features. Alternatively, the terminal device 410 can directly perform feature extraction of the first content, feature extraction of the first title, and splicing of the content features and the title features.

[0308] That is, after the terminal device 410 acquires the first content and the first title, the terminal device 410 can perform feature extraction on the first content to obtain content features, and can perform feature extraction on the first title to obtain title features. Subsequently, the terminal device 410 can send the content features and the title features to the cloud server 420. The cloud server 420 can splice the content features and the title features to obtain spliced features, and can determine relevant features from the feature library according to the spliced features, to determine the candidate topic label corresponding to the first content according to the topic label associated with the relevant features. Alternatively, after the terminal device 410 acquires the first content and the first title, the terminal device 410 can perform feature extraction on the first content to obtain content features, and can perform feature extraction on the first title to obtain title features. Subsequently, the terminal device 410 can splice the content features and the title features to obtain spliced features, and can send the spliced features to the cloud server 420. The cloud server 420 can determine relevant features from the feature library according to the spliced features, and can determine the candidate topic label corresponding to the first content according to the topic label associated with the relevant features.

[0309] It should be understood that the embodiments of the present application do not limit the specific manner in which the terminal device 410 or the cloud server 420 performs feature extraction on the first content, and the specific manner in which the terminal device 410 performs feature extraction on the first title, and the specific manner in which the content features (such as image features or text features, etc.) and the title features are spliced can be determined according to actual application scenarios.

[0310] In one example, for each spliced feature, after determining the similarity between the spliced feature and each preset feature in the feature library, the cloud server 420 can determine a preset feature (such as preset feature A) with a similarity greater than or equal to a certain threshold (such as threshold A) as the relevant feature corresponding to the spliced feature.

[0311] In another example, for each spliced feature, after determining the similarity between the spliced feature and each preset feature in the feature library, the cloud server 420 can determine the N preset features with the largest similarity as the relevant features corresponding to the spliced feature. That is, the N preset features in the feature library that are most similar to the spliced feature can be determined as the relevant features corresponding to the spliced feature.

[0312] It should be noted that the threshold value A can be determined according to an actual application scenario, and embodiments of the present application do not limit the threshold value A. For example, the threshold value A can be determined to be any value such as 0.7, 0.8, or 0.9 according to an actual application scenario. Similarly, the specific value of N can be determined according to an actual application scenario, and embodiments of the present application do not limit the specific value of N. For example, the value of N can be determined to be any value such as 3, 4, or 5 according to an actual application scenario.

[0313] In a possible implementation, when the first content is a video, the content features corresponding to the first content can include image features corresponding to all video frames in the first content. That is, after obtaining the first content and the first title, the cloud server 420 can perform feature extraction on each video frame of the first content to obtain image features corresponding to each video frame.

[0314] In another possible implementation, when the first content is a video, the content features corresponding to the first content can include image features corresponding to part of the video frames in the first video, so as to reduce the number of image features, quickly determine the related features, and then quickly determine the candidate topic label corresponding to the first content according to the topic label associated with the related features, thereby improving the determination speed of the candidate topic label.

[0315] For example, after obtaining the first content, the cloud server 420 or the terminal device 410 can frame the first content to obtain M video frames, so that the cloud server 420 can determine the related features from the feature library according to the image features corresponding to the M video frames and the title features corresponding to the first title.

[0316] It should be noted that the specific value of M can be determined according to an actual application scenario, and embodiments of the present application do not limit the specific value of M. In addition, embodiments of the present application do not limit the specific manner of framing, which can be determined according to an actual application scenario. For example, the first content can be framed by a random extraction manner to obtain M video frames. For example, the first content can be framed by a key frame extraction manner to obtain M video frames. For example, the first content can be segmented by shots, and key frames can be extracted from each segmented shot to obtain M video frames. The key frames extracted after shot segmentation can be one or more frames that are the most important and representative in the shot.

[0317] In one example, the image features corresponding to the first content can include image features corresponding to all video frames in the M video frames. That is, after the M video frames are obtained by frame extraction, the cloud server 420 or the terminal device 410 can perform feature extraction on each of the M video frames to obtain image features corresponding to each of the M video frames. For the image features corresponding to each of the M video frames, the cloud server 420 or the terminal device 410 can splice the image features with the title features to obtain spliced features. Subsequently, the cloud server 420 can determine the relevant features corresponding to the image features corresponding to each of the M video frames according to the similarity between the spliced features and the preset features in the feature library.

[0318] In another example, the image features corresponding to the first content can include image features corresponding to part of the video frames in the M video frames, so as to reduce the number of image features, quickly determine the relevant features, quickly determine the candidate topic label corresponding to the first content according to the topic label associated with the relevant features, and improve the speed of determining the candidate topic label.

[0319] For example, after the M video frames are obtained by frame extraction, the cloud server 420 or the terminal device 410 (hereinafter, the cloud server 420 will be exemplarily described) can perform deduplication processing on the M video frames according to the relevance between the M video frames and the first title, delete irrelevant video frames, for example, delete duplicate video frames and / or transition frames and other irrelevant video frames, obtain the most representative key video frames corresponding to the first content, and splice the image features corresponding to the key video frames with the title features to obtain spliced features, so that the cloud server 420 can determine the relevant features from the feature library only according to the spliced features corresponding to the key video frames, which can reduce the number of spliced features when the cloud server 420 determines the relevant features, improve the speed and efficiency of the cloud server 420 determining the relevant features, and improve the accuracy of determining the relevant features, thereby improving the speed, efficiency and accuracy of determining the candidate topic label, reducing the waiting time of the user, and improving the user experience.

[0320] Specifically, after the M video frames are obtained by frame extraction, the cloud server 420 can perform feature extraction on each of the M video frames to obtain image features corresponding to each of the M video frames. For example, the image features corresponding to each of the M video frames can be denoted as B1, B2, …, BM respectively. M .

[0321] Subsequently, the cloud server 420 can determine B1, B2, …, BM respectively. MThe similarity between the title features corresponding to the first title, and B1, B2, …, Bn can be sorted according to the similarity from large to small, and the sorting result is obtained. Assuming that the sorting result is B1, B2, …, Bn. M The sorting is performed, and the sorting result is obtained. Assuming that the sorting result is B1, B2, …, Bn. M .

[0322] It should be understood that in the scenario of determining the similarity between two features according to the distance, since the smaller the distance is, the greater the similarity is, therefore, the cloud server 420 can also directly sort B1, B2, …, Bn according to the distance from small to large, and the sorting result is obtained. For example, the cosine distance between B1, B2, …, Bn and the title features can be calculated respectively, and B1, B2, …, Bn can be sorted according to the cosine distance from small to large, and the sorting result is obtained. M M The sorting is performed, and the sorting result is obtained. Assuming that the sorting result is B1, B2, …, Bn. M The sorting is performed, and the sorting result is obtained.

[0323] After obtaining the sorting result, the cloud server 420 can determine the first image feature in the sorting result as the image feature in the key feature set corresponding to the first content, and can perform the deduplication processing of the image features according to the similarity between the other image features in the sorting result and each image feature in the key feature set.

[0324] That is, the cloud server 420 can determine B1 as the image feature in the key feature set corresponding to the first content, and can determine the similarity between B2 and each image feature in the key feature set. Since the key feature set at this time only includes B1, the cloud server 420 can determine the similarity between B2 and B1. When it is determined that the similarity between B2 and each image feature in the key feature set is less than a certain threshold (for example, threshold B), that is, when it is determined that the similarity between B2 and B1 is less than the threshold B, the cloud server 420 can determine B2 as the image feature in the key feature set. When it is determined that the similarity between B2 and a certain image feature in the key feature set is greater than or equal to the threshold B, that is, when it is determined that the similarity between B2 and B1 is greater than or equal to the threshold B, the cloud server 420 can determine that there is an image feature similar to B2 in the key feature set, at this time, the cloud server 420 can determine that B2 is not the image feature in the key feature set, so as to reduce the number of similar image features when determining the relevant features. The following will be exemplarily illustrated by taking B2 as the image feature in the key feature set, that is, the key feature set corresponding to the first content can include B1 and B2.

[0325] ​The cloud server 420 can continue to determine the similarity between B3 and each image feature in the key feature set. Since the key feature set at this time includes B1 and B2, the cloud server 420 can determine the similarity between B3 and B1, and the similarity between B3 and B2. When it is determined that the similarity between B3 and B1 is less than the threshold B, and it is determined that the similarity between B3 and B2 is less than the threshold B, the cloud server 420 can determine B3 as an image feature in the key feature set. When it is determined that the similarity between B3 and B1 is greater than or equal to the threshold B, or it is determined that the similarity between B3 and B2 is greater than or equal to the threshold B, the cloud server 420 can determine that B3 is not an image feature in the key feature set. The following will be exemplarily described taking B3 as an image feature in the key feature set.

[0326] Subsequently, the cloud server 420 can continue to determine the similarity between B4 and each image feature in the key feature set. Since the key feature set at this time includes B1 and B2, the cloud server 420 can determine the similarity between B4 and B1, and the similarity between B4 and B2. When it is determined that the similarity between B4 and B1 is less than the threshold B, and it is determined that the similarity between B4 and B2 is less than the threshold B, the cloud server 420 can determine B4 as an image feature in the key feature set. When it is determined that the similarity between B4 and B1 is greater than or equal to the threshold B, or it is determined that the similarity between B4 and B2 is greater than or equal to the threshold B, the cloud server 420 can determine that B4 is not an image feature in the key feature set.

[0327] By analogy, until it is determined that B M is an image feature in the key feature set. After it is determined that B M is an image feature in the key feature set, the cloud server 420 can determine the related feature according to each image feature in the key feature set and the title feature, and each preset feature in the feature library. That is, the cloud server 420 can splice each image feature in the key feature set with the title feature to obtain a spliced feature, and can determine the similarity between the spliced feature and each preset feature in the feature library, to determine the related feature according to the similarity.

[0328] It should be noted that the threshold B can be determined according to the actual application scenario, and the embodiments of the present application do not limit this. For example, the threshold B can be determined as any numerical value such as 0.4 or 0.3 according to the actual application scenario.

[0329] In one example, after determining the relevant features, the cloud server 420 can determine K relevant features with the smallest similarity between the relevant features and the corresponding stitching features according to the similarity, to determine the candidate topic label corresponding to the first content according to the topic label associated with the K relevant features. The K relevant features can be the K preset features most relevant to the first content in the feature library, to determine the candidate topic label corresponding to the first content according to the topic label associated with the K preset features most relevant to the first content, which can improve the speed of determining the candidate topic label.

[0330] It should be noted that the specific value of K can be determined according to the actual application scenario, and the value of K is not limited in the embodiments of the present application.

[0331] In the embodiments of the present application, after determining the relevant features or K relevant features, the cloud server 420 can obtain the topic label associated with each relevant feature, and can determine the candidate topic label corresponding to the first content according to the topic label associated with each relevant feature. Each relevant feature can be associated with one or more topic labels.

[0332] In one example, the cloud server 420 can determine all the topic labels associated with the relevant features as the candidate topic label corresponding to the first content.

[0333] For example, when it is determined that the relevant features include D1 and D2, the topic label associated with D1 includes label E1 and label E2, and the topic label associated with D2 includes label E1, label E3 and label E4, the cloud server 420 can determine label E1, label E2, label E3 and label E4 as the candidate topic label corresponding to the first content.

[0334] In another example, the cloud server 420 can determine part of the topic labels associated with the relevant features as the candidate topic label corresponding to the first content.

[0335] It should be understood that for each relevant feature, the feature library can store the relevant feature, the topic label associated with the relevant feature, and the confidence of each topic label. Alternatively, for each relevant feature, the feature library can store the relevant feature and the ID of the preset content associated with the relevant feature. The preset content can be associated with the topic label and the confidence of each topic label, that is, the ID of the preset content associated with the preset feature can be used to determine the topic label associated with the preset feature and the confidence of each topic label. The confidence can be used to represent the weight of the topic label associated with the corresponding relevant feature. For each relevant feature, the confidence of the topic label associated with the relevant feature can be determined according to the actual application scenario.

[0336] In a possible implementation, for each related feature, when obtaining the topic labels associated with the related feature, the cloud server 420 can obtain the confidence corresponding to each topic label associated with the related feature, and can determine the candidate topic label corresponding to the related feature according to the confidence corresponding to each topic label associated with the related feature. Subsequently, the cloud server 420 can determine the candidate topic label corresponding to the first content according to the candidate topic label corresponding to each related feature.

[0337] For example, for each related feature, the cloud server 420 can determine, as the candidate topic label corresponding to the related feature, the topic label with a confidence greater than or equal to a threshold value (for example, threshold value C). It should be understood that the specific value of the threshold value C can be determined according to an actual application scenario, and the embodiments of the present application do not limit this.

[0338] For example, in a case where it is determined that the related features include D1 and D2, the topic labels associated with D1 include label E1, label E2 and label E3, the topic labels associated with D2 include label E1, label E4 and label E5, in D1, the confidence corresponding to label E1 is 0.5, the confidence corresponding to label E2 is 0.4, and the confidence corresponding to label E3 is 0.1. In D2, the confidence corresponding to label E1 is 0.3, the confidence corresponding to label E4 is 0.4, and the confidence corresponding to label E5 is 0.3.

[0339] Suppose that the threshold value C is 0.2, the cloud server 420 can determine that the candidate topic labels corresponding to D1 include label E1 and label E2, the candidate topic labels corresponding to D2 include label E1, label E4 and label E5, and can determine the candidate topic label corresponding to the first content according to label E1, label E2, label E4 and label E5.

[0340] For example, for each related feature, the cloud server 420 can determine, as the candidate topic label corresponding to the related feature, the W topic labels with the largest confidence. It should be understood that the specific value of W can be determined according to an actual application scenario, and the embodiments of the present application do not limit this.

[0341] In another possible implementation, for each related feature, the cloud server 420 can obtain the confidence degrees corresponding to the topic labels associated with the related feature when obtaining the topic labels associated with the related feature. After obtaining the confidence degrees corresponding to the topic labels associated with each related feature, for each topic label, the cloud server 420 can determine the final confidence degree of the topic label according to the similarities corresponding to each related feature (i.e., the similarity between the related feature and the corresponding stitching feature) and the confidence degrees corresponding to the topic label. Subsequently, the cloud server 420 can determine the candidate topic labels according to the final confidence degrees of the topic labels, and can determine the candidate topic label corresponding to the first content according to the candidate topic labels.

[0342] For example, the image features in the key feature set corresponding to the first content include B1 and B2, the related features determined according to B1 include D1 and D2, the related features determined according to B2 include D3 and D4, the similarity between D1 and the stitching feature corresponding to B1 is 1.0, the similarity between D2 and the stitching feature corresponding to B1 is 1.0, the similarity between D3 and the stitching feature corresponding to B2 is 0.8, and the similarity between D4 and the stitching feature corresponding to B2 is 0.8. The topic labels associated with D1 include label E1 and label E2, the confidence degree corresponding to label E1 is 0.8, and the confidence degree corresponding to label E2 is 0.2. The topic labels associated with D2 include label E1 and label E3, the confidence degree corresponding to label E1 is 0.7, and the confidence degree corresponding to label E3 is 0.3. The topic labels associated with D3 include label E3 and label E4, the confidence degree corresponding to label E3 is 0.8, and the confidence degree corresponding to label E4 is 0.2. The topic labels associated with D4 include label E2 and label E4, the confidence degree corresponding to label E2 is 0.6, and the confidence degree corresponding to label E4 is 0.4.

[0343] At this time, the cloud server 420 can determine that the topic labels associated with B1 include label E1, label E2 and label E3, and can determine that the confidence degrees corresponding to label E1 include 0.8 and 0.7, the confidence degree corresponding to label E2 includes 0.2, and the confidence degree corresponding to label E3 includes 0.3. For label E1 associated with B1, since label E1 is the topic label associated with D1 and D2, the cloud server 420 can combine the confidence degrees (i.e., 0.8 and 0.7) corresponding to label E1 according to the similarity between D1 and the stitching feature corresponding to B1 (i.e., 1.0) and the similarity between D2 and the stitching feature corresponding to B1 (i.e., 1.0), to obtain the combined confidence degree of label E1. That is, the combined confidence degree of label E1 = 1.0*0.8+1.0*0.7 = 1.5.

[0344] For the label E2 associated with B1, since the label E2 is a topic label associated with D1, the cloud server 420 can merge the confidence corresponding to the label E2 (i.e., 0.2) according to the similarity between the splicing features corresponding to D1 and B1 (i.e., 1.0), to obtain the merged confidence of the label E2. That is, the merged confidence of the label E2 = 1.0*0.2 = 0.2.

[0345] For the label E3 associated with B1, since the label E3 is a topic label associated with D2, the cloud server 420 can merge the confidence corresponding to the label E3 (i.e., 0.3) according to the similarity between the splicing features corresponding to D2 and B1 (i.e., 1.0), to obtain the merged confidence of the label E3. That is, the merged confidence of the label E3 = 1.0*0.3 = 0.3.

[0346] Subsequently, for the labels E1, E2 and E3 associated with B1, after obtaining the merged confidence of the label E1, the merged confidence of the label E2 and the merged confidence of the label E3, the cloud server 420 can normalize the merged confidence to obtain the normalized confidence of the label E1, the normalized confidence of the label E2 and the normalized confidence of the label E3. For example, the normalized confidence of the label E1 = 1.5 / (1.5+0.2+0.3) = 0.75, the normalized confidence of the label E2 = 0.2 / (1.5+0.2+0.3) = 0.1, and the normalized confidence of the label E3 = 0.3 / (1.5+0.2+0.3) = 0.15.

[0347] Similarly, the cloud server 420 can determine that the topic labels associated with B2 include labels E2, E3 and E4, and can determine that the confidence corresponding to the label E2 includes 0.6, the confidence corresponding to the label E3 includes 0.8, and the confidence corresponding to the label E4 includes 0.2 and 0.4. For the label E2 associated with B2, since the label E2 is a topic label associated with D4, the cloud server 420 can merge the confidence corresponding to the label E2 (i.e., 0.6) according to the similarity between the splicing features corresponding to D4 and B2 (i.e., 0.8), to obtain the merged confidence of the label E2. That is, the merged confidence of the label E2 = 0.8*0.6 = 0.48.

[0348] For the label E3 associated with B2, since the label E3 is a topic label associated with D3, the cloud server 420 can merge the confidence corresponding to the label E3 (i.e., 0.8) according to the similarity between the splicing features corresponding to D3 and B2 (i.e., 0.8), to obtain the merged confidence of the label E3. That is, the merged confidence of the label E3 = 0.8*0.8 = 0.64.

[0349] For the label E4 associated with B2, since the label E3 is the topic label associated with D3 and D4 respectively, the cloud server 420 can combine the confidence corresponding to the label E4 (i.e., 0.2 and 0.4) according to the similarity between the stitching features corresponding to D3 and B2 (i.e., 0.8) and the similarity between the stitching features corresponding to D4 and B2 (i.e., 0.8), to obtain the combined confidence of the label E4. That is, the combined confidence of the label E4 = 0.8*0.2 + 0.8*0.4 = 0.48.

[0350] Subsequently, for the label E2, the label E3 and the label E4 associated with B2, after obtaining the combined confidence of the label E2, the combined confidence of the label E3 and the combined confidence of the label E4, the cloud server 420 can normalize the combined confidence to obtain the normalized confidence of the label E2, the normalized confidence of the label E3 and the normalized confidence of the label E4. For example, the normalized confidence of the label E2 = 0.48 / (0.48 + 0.64 + 0.48) = 0.3, the normalized confidence of the label E3 = 0.64 / (0.48 + 0.64 + 0.48) = 0.4, and the normalized confidence of the label E4 = 0.48 / (0.48 + 0.64 + 0.48) = 0.3.

[0351] Wherein, after determining the normalized confidence of each topic label associated with B1 (which can be referred to as confidence B1 for ease of understanding) and the normalized confidence of each topic label associated with B2 (which can be referred to as confidence B2 for ease of understanding), for each topic label, the cloud server 420 can combine the confidence B1 and the confidence B2 corresponding to the topic label to obtain the final confidence of the topic label. That is, the final confidence of the label E1 = 0.75, the final confidence of the label E2 = 0.1 + 0.3 = 0.4, the final confidence of the label E3 = 0.15 + 0.4 = 0.55, and the final confidence of the label E4 = 0.3.

[0352] It should be understood that after obtaining the final confidence of each topic label, the cloud server 420 can determine the candidate topic label according to the final confidence of each topic label.

[0353] In one example, the cloud server 420 can determine the topic label with the final confidence greater than or equal to a certain threshold (for example, threshold D) as the candidate topic label. Wherein, the threshold D can be determined according to the actual application scenario, and the embodiments of the present application do not limit this. For example, the threshold D can be determined as 0.4, 0.5 or 0.6 or any other value according to the actual application scenario.

[0354] For example, in a specific scenario, the threshold value D can be determined as 0.5, and when the final confidence of the label E1 is determined as 0.75, the final confidence of the label E2 is determined as 0.4, the final confidence of the label E3 is determined as 0.55, and the final confidence of the label E4 is determined as 0.3, the cloud server 420 can determine the label E1 and the label E3 as the candidate topic labels.

[0355] In another example, the cloud server 420 can determine R topic labels with the largest final confidence as the candidate topic labels. The specific value of R can be determined according to an actual application scenario, which is not limited in the embodiments of the present application. For example, the specific value of R can be determined as 3 or 4 or any other value according to an actual application scenario.

[0356] For example, in a specific scenario, the value of R can be determined as 3, and when the final confidence of the label E1 is determined as 0.75, the final confidence of the label E2 is determined as 0.4, the final confidence of the label E3 is determined as 0.55, and the final confidence of the label E4 is determined as 0.3, the cloud server 420 can determine the label E1, the label E3 and the label E2 as the candidate topic labels.

[0357] In a possible implementation, after determining the candidate topic labels, the cloud server 420 can directly determine all the candidate topic labels as the candidate topic labels corresponding to the first content. For example, in a specific scenario, when the candidate topic labels include the label E1, the label E2 and the label E3, the cloud server 420 can directly determine the label E1, the label E2 and the label E3 as the candidate topic labels corresponding to the first content.

[0358] In another possible implementation, after determining the candidate topic labels, the cloud server 420 can determine the candidate topic labels corresponding to the first content according to the confidence of each candidate topic label. For example, the candidate topic label with a confidence greater than or equal to a threshold value (for example, the threshold value E) can be determined as the candidate topic label corresponding to the first content. For example, the S candidate topic labels with the largest confidence can be determined as the candidate topic labels corresponding to the first content. It should be understood that the specific value of the threshold value E and the specific value of S can be determined according to an actual application scenario, which is not limited in the embodiments of the present application.

[0359] The process of determining the hierarchical relationship between the candidate topic labels corresponding to the first content by the cloud server 420 or the terminal device 410 will be described in detail below.

[0360] In one example, after determining the candidate topic labels corresponding to the first content, the cloud server 420 or the terminal device 410 can determine the hierarchical relationship between the candidate topic labels. The hierarchical relationship between the candidate topic labels can include two levels, or can include three or more levels. The following describes two cases: (1) the hierarchical relationship between the candidate topic labels includes two levels; and (2) the hierarchical relationship between the candidate topic labels includes three or more levels.

[0361] (1) The hierarchical relationship between the candidate topic labels includes two levels.

[0362] In one possible implementation, after determining the candidate topic labels corresponding to the first content, the cloud server 420 can obtain the co-occurrence relationship of the candidate topic labels, and can divide the candidate topic labels into one or more clusters according to the co-occurrence relationship of the candidate topic labels. For each cluster, the cloud server 420 can sort the candidate topic labels in the cluster to obtain a sorting result corresponding to the cluster, and can determine the first candidate topic label in the sorting result as the first level candidate topic label corresponding to the cluster, and determine the other candidate topic labels as the second level candidate topic labels corresponding to the cluster.

[0363] In another possible implementation, after determining the candidate topic labels corresponding to the first content, the cloud server 420 can obtain the co-occurrence relationship of the candidate topic labels, and can send the co-occurrence relationship of the candidate topic labels to the terminal device 410. The terminal device 410 can divide the candidate topic labels into one or more clusters according to the co-occurrence relationship of the candidate topic labels. For each cluster, the terminal device 410 can sort the candidate topic labels in the cluster to obtain a sorting result corresponding to the cluster, and can determine the first candidate topic label in the sorting result as the first level candidate topic label corresponding to the cluster, and determine the other candidate topic labels as the second level candidate topic labels corresponding to the cluster.

[0364] For example, for each cluster, the cloud server 420 or the terminal device 410 can sort the candidate topic labels in the cluster according to the order from high to low of the confidence to obtain a sorting result corresponding to the cluster. That is, for each cluster, the first candidate topic label in the sorting result can be the candidate topic label with the highest confidence in the cluster. The last candidate topic label in the sorting result can be the candidate topic label with the lowest confidence in the cluster.

[0365] For example, the cloud server 420 or the terminal device 410 can obtain the hotness corresponding to each candidate topic label. For each cluster, the cloud server 420 or the terminal device 410 can sort the candidate topic labels in the cluster according to the order of the hotness corresponding to each candidate topic label from high to low, to obtain the sorting result corresponding to the cluster. That is, the first candidate topic label in the sorting result can be the candidate topic label with the highest hotness in the cluster. The last candidate topic label in the sorting result can be the candidate topic label with the lowest hotness in the cluster.

[0366] For example, for each cluster, the cloud server 420 or the terminal device 410 can calculate the score corresponding to each candidate topic label according to the hotness and the confidence corresponding to each candidate topic label in the cluster, and can sort the candidate topic labels in the cluster according to the order of the score from high to low, to obtain the sorting result corresponding to the cluster. That is, the first candidate topic label in the sorting result can be the candidate topic label with the highest score in the cluster. The last candidate topic label in the sorting result can be the candidate topic label with the lowest score in the cluster.

[0367] For example, for each candidate topic label, the score corresponding to the candidate topic label can be the product of the hotness corresponding to the candidate topic label and the confidence corresponding to the candidate topic label.

[0368] It should be understood that the co-occurrence relationship of the candidate topic labels can refer to the relationship that two or more candidate topic labels appear at the same time. For example, it can refer to the relationship that two or more candidate topic labels are associated with the same preset feature. The cloud server 420 can determine the co-occurrence relationship of the candidate topic labels by analyzing the topic labels associated with each preset feature in the feature library, thereby determining the co-occurrence relationship of the candidate topic labels.

[0369] It should be noted that the hotness corresponding to the candidate topic label can be determined according to the number of uses and / or the number of views of the candidate topic label within a preset time. For example, the hotness corresponding to each candidate topic label can be determined according to the number of uses and the number of views of each candidate topic label in the past week or month. The more the number of uses of the candidate topic label, the higher the hotness of the candidate topic label; the less the number of uses of the candidate topic label, the lower the hotness of the candidate topic label. The more the number of views of the candidate topic label, the higher the hotness of the candidate topic label; the less the number of views of the candidate topic label, the lower the hotness of the candidate topic label.

[0370] In another possible implementation, after determining the candidate topic label corresponding to the first content, the cloud server 420 or the terminal device 410 (for ease of understanding, the terminal device 410 is exemplarily taken as an example below) can sort all candidate topic labels according to the confidence and / or the heat, to obtain a sorting result. Subsequently, the terminal device 410 can take the first candidate topic label in the sorting result as a root node, for example, as a first root node. The specific content of sorting all candidate topic labels according to the confidence and / or the heat can refer to the foregoing related content of sorting the candidate topic labels of a certain cluster according to the confidence and / or the heat, and for the sake of brevity, will not be repeated here.

[0371] For the second candidate topic label in the sorting result, the terminal device 410 can determine the semantic similarity between the second candidate topic label and the first root node. When the semantic similarity between the second candidate topic label and the first root node is greater than or equal to a certain threshold (for example, threshold F), the terminal device 410 can determine the second candidate topic label as a child node of the first root node. When the semantic similarity between the second candidate topic label and the first root node is less than the threshold F, the terminal device 410 can determine the second candidate topic label as a new root node, for example, as a second root node.

[0372] For the third candidate topic label in the sorting result, the terminal device 410 can in turn determine the semantic similarity between the third candidate topic label and each root node. When the semantic similarity between the third candidate topic label and a certain root node is greater than or equal to the threshold F, the terminal device 410 can determine the third candidate topic label as a child node of the root node. When the semantic similarity between the third candidate topic label and all root nodes is less than the threshold F, the terminal device 410 can determine the third candidate topic label as a new root node.

[0373] For example, when the second candidate topic label is determined as the second root node, for the third candidate topic label, the terminal device 410 can determine the semantic similarity between the third candidate topic label and the first root node. When the semantic similarity between the third candidate topic label and the first root node is greater than or equal to the threshold value F, the terminal device 410 can determine the third candidate topic label as a child node of the first root node. When the semantic similarity between the third candidate topic label and the first root node is less than the threshold value F, the terminal device 410 can determine the semantic similarity between the third candidate topic label and the second root node. When the semantic similarity between the third candidate topic label and the second root node is greater than or equal to the threshold value F, the terminal device 410 can determine the third candidate topic label as a child node of the second root node. When the semantic similarity between the third candidate topic label and the second root node is less than the threshold value F, the terminal device 410 can determine the third candidate topic label as a new root node, for example, as the third root node.

[0374] Similarly, for the fourth candidate topic label in the sorting result, the terminal device 410 can in turn determine the semantic similarity between the fourth candidate topic label and each root node. When the semantic similarity between the fourth candidate topic label and a certain root node is greater than or equal to the threshold value F, the terminal device 410 can determine the fourth candidate topic label as a child node of the root node. When the semantic similarity between the fourth candidate topic label and all root nodes is less than the threshold value F, the terminal device 410 can determine the fourth candidate topic label as a new root node.

[0375] By analogy, until the determination of the last candidate topic label in the sorting result is completed. After the determination of the last candidate topic label is completed, the terminal device 410 can determine the hierarchical relationship between the candidate topic labels according to each root node and the child nodes of each root node. For example, the terminal device 410 can determine each root node (i.e., the candidate topic label indicated by the root node) as a candidate topic label of the first level, and can determine each child node (i.e., the candidate topic label indicated by the child node) as a candidate topic label of the second level under the corresponding root node.

[0376] It should be noted that the specific value of the threshold value F can be determined according to the actual application scenario, and the application embodiment does not make specific limitations on the value of the threshold value F.

[0377] For example, please refer to FIG. 6, which shows an example diagram one of the hierarchical relationship corresponding to the candidate topic label provided by the application embodiment.

[0378] When the candidate topic labels corresponding to the first content include label El, label E2, label E3, label E4, label E5, label E6, label E7, label E8, label E9 and label E10, and the sorting result is: label E10 -- label E9 -- label E8 -- label E7 -- label E6 -- label E5 -- label E4 -- label E3 -- label E2 -- label El, it is assumed that the terminal device 410 determines label E10 as the first root node, label E8 as the second root node, label E5 as the third root node, label E3 as the fourth root node, label E9 as a child node of the first root node, label E7 as a child node of the first root node, label E6 as a child node of the second root node, label E4 as a child node of the third root node, label E2 as a child node of the second root node, and label El as a child node of the fourth root node according to the above manner, the terminal device 410 can determine label E10, label E8, label E5 and label E3 as the candidate topic labels of the first level, and can determine label E9 and label E7 as the candidate topic labels of the second level under label E10, label E6 and label E2 as the candidate topic labels of the second level under label E8, label E4 as the candidate topic label of the second level under label E5, and label El as the candidate topic label of the second level under label E3. That is, the hierarchical relationship among the candidate topic labels corresponding to the first content can be as shown in FIG. 6.

[0379] (ii) The hierarchical relationship among the candidate topic labels includes three or more levels.

[0380] In a possible implementation, after determining the candidate topic labels corresponding to the first content, the cloud server 420 or the terminal device 410 (for ease of understanding, the cloud server 420 is exemplarily taken as an example in the following) can divide the candidate topic labels into one or more clusters according to the co-occurrence relationship of the candidate topic labels. For each cluster (for example, taking cluster A as an example), the cloud server 420 can sort the candidate topic labels of the cluster A according to the confidence and / or the heat, to obtain a sorting result (for example, sorting result A) corresponding to the cluster A. Subsequently, the cloud server 420 can take the first candidate topic label in the sorting result A as a root node (for example, root node A) of the cluster A, and can take the second candidate topic label in the sorting result A as a child node of the root node A, for example, as the first child node of the root node A. The specific content of the co-occurrence relationship can be referred to the related content of the co-occurrence relationship described above, and the specific content of sorting the candidate topic labels of the cluster A according to the confidence and / or the heat can be referred to the related content described above, which will not be described herein again.

[0381] For the third candidate topic label in the ranking result A, the cloud server 420 can determine the semantic similarity between the third candidate topic label and the first child node of the root node A. When the semantic similarity between the third candidate topic label and the first child node of the root node A is greater than or equal to a certain threshold (for example, threshold G), the cloud server 420 can determine the third candidate topic label as a child node of the first child node of the root node A. When the semantic similarity between the third candidate topic label and the first child node of the root node A is less than the threshold G, the cloud server 420 can determine the third candidate topic label as a new child node of the root node A, for example, as the second child node of the root node A.

[0382] For the fourth candidate topic label in the ranking result A, the cloud server 420 can in turn determine the semantic similarity between the fourth candidate topic label and each child node of the root node A. When the semantic similarity between the fourth candidate topic label and all child nodes of the root node A is less than the threshold G, the cloud server 420 can determine the fourth candidate topic label as a new child node of the root node A (i.e., as a leaf node of the root node A). When the semantic similarity between the fourth candidate topic label and a certain child node (for example, the first child node) of the root node A is greater than or equal to the threshold G, the cloud server 420 can determine whether the first child node has a child node. When the first child node has no child node, the cloud server 420 can determine the fourth candidate topic label as a child node of the first child node (i.e., as a leaf node of the first child node). When the first child node has a child node, the cloud server 420 can in turn determine the semantic similarity between the fourth candidate topic label and each child node of the first child node. When the semantic similarity between the fourth candidate topic label and all child nodes under the first child node is less than a certain threshold (for example, threshold H), the cloud server 420 can determine the fourth candidate topic label as a new child node of the first child node (i.e., as a leaf node of the first child node). When the semantic similarity between the fourth candidate topic label and a certain child node (for example, child node a) of the first child node is greater than or equal to the threshold H, the cloud server 420 can determine whether the child node a has a child node. When the child node a has no child node, the cloud server 420 can determine the fourth candidate topic label as a child node of the child node a (i.e., as a leaf node of the child node a). When the child node a has a child node, the cloud server 420 can in turn determine the semantic similarity between the fourth candidate topic label and each child node of the child node a, and can determine the parent node corresponding to the fourth candidate topic label according to the semantic similarity and a certain threshold (for example, threshold I), and so on, until the fourth candidate topic label is determined as a leaf node.

[0383] Similarly, until the determination of the last candidate topic label in the ranking result corresponding to the cluster A is completed. Wherein, after the determination of the last candidate topic label in the ranking result corresponding to all clusters is completed, the cloud server 420 can determine the hierarchical relationship between the candidate topic labels according to the root nodes of each cluster and the child nodes of each root node.

[0384] For example, please refer to FIG. 7, which shows an example diagram two of the hierarchical relationship between the candidate topic labels provided by the embodiments of the present application.

[0385] When the candidate topic labels corresponding to the first content include labels E1, E2, E3, E4, E5, E6, E7, E8, E9, E10, E11 and E12, it is assumed that the cloud server 420 can determine labels E1, E2, E3, E4, E5 and E6 as a cluster (for example, cluster A) and determine labels E7, E8, E9, E10, E11 and E12 as another cluster (for example, cluster B) according to the common relationship.

[0386] It is assumed that the ranking result corresponding to the cluster A is: label E6--label E5--label E4--label E3--label E2--label E1, at this time, the cloud server 420 can determine label E6 as the root node of the cluster A, and can determine label E5 as the first child node of the root node (i.e. label E6).

[0387] For the third candidate topic label (i.e. label E4) in the ranking result corresponding to the cluster A, the cloud server 420 can determine the semantic similarity between label E4 and label E5. It is assumed that the semantic similarity between label E4 and label E5 is less than the threshold G, the cloud server 420 can determine label E4 as the child node of the root node (i.e. label E6), that is, as the second child node of label E6.

[0388] For the fourth candidate topic label (i.e. label E3) in the ranking result corresponding to the cluster A, the cloud server 420 can determine the semantic similarity between label E3 and label E5. It is assumed that the semantic similarity between label E3 and label E5 is less than the threshold G, the cloud server 420 can determine the semantic similarity between label E3 and label E4. It is assumed that the semantic similarity between label E3 and label E4 is greater than or equal to the threshold H, the cloud server 420 can determine label E3 as the child node of the second child node (i.e. label E4).

[0389] For the fifth candidate topic label in the ranking result corresponding to cluster A (i.e., label E2), the cloud server 420 can determine the semantic similarity between label E2 and label E5. Assuming that the semantic similarity between label E2 and label E5 is less than threshold G, the cloud server 420 can determine the semantic similarity between label E2 and label E4. Assuming that the semantic similarity between label E2 and label E4 is greater than or equal to threshold H, the cloud server 420 can determine the semantic similarity between label E2 and label E3. Assuming that the semantic similarity between label E2 and label E3 is less than threshold I, the cloud server 420 can determine label E2 as a child node of the second child node (i.e., label E4), and label E2 and label E3 can be sibling nodes under the second child node.

[0390] For the last candidate topic label in the ranking result corresponding to cluster A (i.e., label E1), the cloud server 420 can determine the semantic similarity between label E1 and label E5. Assuming that the semantic similarity between label E1 and label E5 is less than threshold G, the cloud server 420 can determine the semantic similarity between label E1 and label E4. Assuming that the semantic similarity between label E1 and label E4 is greater than or equal to threshold H, the cloud server 420 can determine the semantic similarity between label E1 and label E3. Assuming that the semantic similarity between label E1 and label E3 is greater than or equal to threshold I, the cloud server 420 can determine label E1 as a child node of label E3.

[0391] Similarly, assuming that the ranking result corresponding to cluster B is: label E7 -- label E8 -- label E9 -- label E10 -- label E11 -- label E12, the cloud server 420 can determine label E7 as the root node of cluster B, and can determine label E8 as the first child node of the root node (i.e., label E7). Assuming that it is determined according to the above manner that: label E9 is a child node of the first child node (i.e., label E8), label E10 is the second child node of the root node, label E11 is a child node of the first child node, and label E12 is a child node of the second child node (i.e., label E10).

[0392] Therefore, the cloud server 420 can determine, according to the root node and the child node of the cluster A and the root node and the child node of the cluster B, that the label E6 and the label E7 are candidate topic labels of the first level, the label E5 and the label E4 are candidate topic labels of the second level under the label E6, the label E8 and the label E10 are candidate topic labels of the second level under the label E7, the label E3 and the label E2 are candidate topic labels of the third level under the label E4, the label E9 and the label E11 are candidate topic labels of the third level under the label E8, the label E12 is a candidate topic label of the third level under the label E10, and the label E1 is a candidate topic label of the fourth level under the label E3. That is, the hierarchical relationship between the candidate topic labels corresponding to the first content can be as shown in FIG. 7.

[0393] It should be noted that the specific values of the threshold value G, the threshold value H, and the threshold value I and the like can be determined according to actual scenes, and the embodiments of the present application are not limited.

[0394] In another possible implementation, after determining the candidate topic labels corresponding to the first content, the cloud server 420 can sort all the candidate topic labels according to the confidence and / or the heat, obtain a sorting result, and can take the first candidate topic label in the sorting result as a root node, for example, as a first root node.

[0395] For the second candidate topic label in the sorting result, the cloud server 420 can determine the semantic similarity between the second candidate topic label and the first root node. When the semantic similarity between the second candidate topic label and the first root node satisfies greater than or equal to a certain threshold value (for example, the threshold value J), the cloud server 420 can determine the second candidate topic label as a child node of the first root node. When the semantic similarity between the second candidate topic label and the first root node is less than the threshold value J, the cloud server 420 can determine the second candidate topic label as a new root node, for example, as a second root node.

[0396] For the third candidate topic label in the sorting result, the cloud server 420 can determine the semantic similarity between the third candidate topic label and each root node in turn. When the semantic similarity between the third candidate topic label and all root nodes is less than the threshold J, the cloud server 420 can determine the third candidate topic label as a new root node. When the semantic similarity between the third candidate topic label and a root node (for example, the first root node) is greater than or equal to the threshold J, the cloud server 420 can determine whether the first root node has a child node. When the first root node has no child node, the cloud server 420 can determine the third candidate topic label as a child node (i.e., a leaf node) of the first root node. When the first root node has a child node, the cloud server 420 can determine the semantic similarity between the third candidate topic label and each child node of the first root node in turn. When the semantic similarity between the third candidate topic label and all child nodes of the first root node is less than a certain threshold (for example, the threshold K), the cloud server 420 can determine the third candidate topic label as a new child node of the first root node, i.e., as a leaf node of the first root node. When the semantic similarity between the third candidate topic label and a child node of the first root node is greater than or equal to the threshold K, the cloud server 420 can determine whether the node has a child node, and can determine the parent node corresponding to the third candidate topic label according to whether the child node has a child node, until the third candidate topic label is determined as a leaf node.

[0397] By analogy, until the determination of the last candidate topic label in the sorting result is completed. After the determination of the last candidate topic label in the sorting result is completed, the cloud server 420 can determine the hierarchical relationship between the candidate topic labels according to the root nodes and the child nodes of the root nodes.

[0398] It should be noted that the specific values of the threshold J and the threshold K, etc. can be determined according to actual scenarios, and the embodiments of the present application are not limited.

[0399] In some embodiments, after the cloud server 420 determines the hierarchical relationship between the candidate topic labels corresponding to the first content, the cloud server 420 can send the candidate topic labels corresponding to the first content and the hierarchical relationship between the candidate topic labels to the terminal device 410. Alternatively, after the cloud server 420 determines the candidate topic labels corresponding to the first content, the cloud server 420 can send the candidate topic labels corresponding to the first content to the terminal device 410. The terminal device 410 can determine the hierarchical relationship between the candidate topic labels. After the terminal device 410 obtains the candidate topic labels corresponding to the first content and the hierarchical relationship between the candidate topic labels, the terminal device 410 can display the candidate topic labels corresponding to the first content in a hierarchical manner according to the hierarchical relationship between the candidate topic labels, so that the user can quickly determine the topic label (which can be referred to as a target topic label below) corresponding to the first content according to the hierarchical display of the candidate topic labels, and the speed and efficiency of the user selecting the target topic label can be improved.

[0400] In one example, when performing hierarchical display of the candidate topic labels, the terminal device 410 can display the candidate topic labels of the first level by default, and can display the candidate topic labels of the second level corresponding to a certain candidate topic label of the first level when detecting a specified operation on the candidate topic label. The specified operation can be a click operation, a touch operation, or a long press operation, etc., which can be determined according to actual application scenarios.

[0401] In one example, when displaying the candidate topic labels of the first level, the terminal device 410 can display the candidate topic labels of the first level according to a preset display order. Similarly, when displaying the candidate topic labels of the second level corresponding to a certain candidate topic label, the terminal device 410 can also display the candidate topic labels of the second level according to the preset display order. Similarly, when displaying the candidate topic labels of the third level corresponding to a certain candidate topic label, the terminal device 410 can also display the candidate topic labels of the third level according to the preset display order, and so on.

[0402] It should be noted that the preset display order can be determined according to actual application scenarios. For example, the preset display order can be determined according to the confidence and / or heat of each candidate topic label. When the preset display order is determined according to the confidence, the higher the confidence, the higher the display order, and the lower the confidence, the lower the display order. Similarly, when the preset display order is determined according to the heat, the higher the heat, the higher the display order, and the lower the heat, the lower the display order. When the preset display order is determined according to the confidence and the heat, the higher the score corresponding to the confidence and the heat, the higher the display order, and the lower the score corresponding to the confidence and the heat, the lower the display order. It should be understood that the score corresponding to the confidence and the heat can be the product of the confidence and the heat.

[0403] In a possible implementation, when the terminal device 410 displays the candidate topic labels corresponding to the first content in a hierarchical manner according to the hierarchical relationship between the candidate topic labels, the user can select a target topic label for the first content from the candidate topic labels corresponding to the first content. Alternatively, the user can also customize a target topic label for the first content. After the selection and / or customization of the target topic label for the first content is completed, the terminal device 410 can obtain the target topic label corresponding to the first content (for example, one or more selected by the user from the candidate topic labels, and / or one or more topic labels customized by the user), and can send the target topic label corresponding to the first content to the cloud server 420. After the cloud server 420 obtains the target topic label corresponding to the first content, the cloud server 420 can save the target topic label corresponding to the first content and the content features corresponding to the first content in association, and update the feature library according to the content features corresponding to the first content and the topic label selected or customized by the user for the first content, which can improve the accuracy of the feature library, and thus can improve the accuracy of the candidate topic labels.

[0404] In an example, for each content feature corresponding to the first content, the cloud server 420 can determine the topic label corresponding to the content feature according to the target topic label corresponding to the first content and the topic label associated with the related feature corresponding to the content feature, and can save the content feature and the topic label corresponding to the content feature in association in the feature library. The related feature corresponding to the content feature can be a related feature determined from the feature library according to the spliced feature corresponding to the content feature (i.e., the spliced feature obtained by splicing the content feature and the title feature).

[0405] For example, when the target topic label corresponding to the first content includes one or more topic labels customized by the user, for each content feature corresponding to the first content, the cloud server 420 can determine the one or more topic labels customized by the user as the topic label corresponding to the content feature.

[0406] For example, when the target topic label corresponding to the first content includes one or more topic labels selected by the user from the candidate topic labels, for each content feature corresponding to the first content, the cloud server 420 can determine the topic label selected by the user as the target topic label from the topic labels associated with the related feature corresponding to the content feature, and can determine the one or more topic labels selected by the user as the target topic label as the topic label corresponding to the content feature.

[0407] For example, the content features corresponding to the first content can include content feature a1 and content feature a2, the related features corresponding to content feature a1 can include D1, and the topic labels associated with D1 can include label E1 and label E2, the related features corresponding to content feature a2 can include D2, and the topic labels associated with D2 can include label E3 and label E4. Assuming that the target topic labels selected by the user for the first content include label E1 and label E4, and in addition, the user also manually customizes label E5 for the first content. At this time, the cloud server 420 can determine label E1 and label E5 as the topic labels corresponding to content feature a1, and can determine label E4 and E5 as the topic labels corresponding to content feature a2. That is, the cloud server 420 can associate and save content feature a1 with label E1 and label E5 to the feature library, and can associate and save content feature a2 with label E4 and label E5 to the feature library.

[0408] In one example, after obtaining the target topic labels corresponding to the first content, the cloud server 420 can adjust the confidence of the topic labels associated with the related features corresponding to the first content according to the target topic labels corresponding to the first content, for example, can increase the confidence corresponding to the topic labels selected by the user, can reduce the confidence corresponding to the topic labels not selected by the user, and improve the accuracy between the topic labels and the preset features in the feature library, so as to improve the accuracy of the candidate topic labels.

[0409] For example, the content features corresponding to the first content can include content feature a1 and content feature a2, the related features corresponding to content feature a1 can include D1, and the topic labels associated with D1 can include label E1 and label E2, the related features corresponding to content feature a2 can include D2, and the topic labels associated with D2 can include label E3 and label E4. Assuming that the target topic labels selected by the user for the first content include label E1 and label E4, that is, label E2 corresponding to related feature D1 and label E3 corresponding to related feature D2 are not selected by the user, at this time, the cloud server 420 can adjust the confidence corresponding to label E1 and label E2 associated with related feature D1, for example, can increase the confidence corresponding to label E1 in related feature D1, and can reduce the confidence corresponding to label E2. Similarly, the cloud server 420 can adjust the confidence corresponding to label E3 and label E4 associated with related feature D2, for example, can increase the confidence corresponding to label E4 in related feature D2, and can reduce the confidence corresponding to label E3.

[0410] The topic label determination method provided by the embodiments of the present application will be described below by taking a video as an example.

[0411] Please refer to FIG. 8, which shows a schematic flowchart of a topic label determination method according to an embodiment of the present application.

[0412] As shown in FIG. 8, when a video is published, a user can select a desired video (e.g., a first video) to be published and can edit a title (e.g., a first title) corresponding to the first video. The terminal device 410 can obtain the first video selected by the user and the first title edited by the user, and can send the first video and the first title to the cloud server 420.

[0413] After obtaining the first video and the first title, the cloud server 420 can perform feature extraction on the first title to obtain title features corresponding to the first title. In addition, the cloud server 420 can also perform frame extraction on the first video to obtain M video frames, and can perform feature extraction on each of the M video frames to obtain image features (e.g., image features A) corresponding to each video frame. Subsequently, the cloud server 420 can perform deduplication on the image features A according to the similarity between each image feature A and the title features to obtain image features B. The image features B can be all or part of the image features A.

[0414] After obtaining the image features B, for each image feature B, the cloud server 420 can concatenate the image feature B with the title features to obtain concatenated features, and can perform feature library retrieval according to the concatenated features to determine related features (e.g., related features C) similar to the concatenated features from the feature library. Subsequently, the cloud server 420 can obtain topic labels associated with each related feature C, and can merge the topic labels associated with each related feature C to determine candidate topic labels corresponding to the first video.

[0415] After determining the candidate topic labels corresponding to the first video, the cloud server 420 can determine the hierarchical relationship between the candidate topic labels according to the commonality relationship of the candidate topic labels or the similarity between the candidate topic labels. Subsequently, the cloud server 420 can send the candidate topic labels corresponding to the first video and the hierarchical relationship between the candidate topic labels to the terminal device 410.

[0416] After obtaining the candidate topic labels corresponding to the first video and the hierarchical relationship between the candidate topic labels, the terminal device 410 can display the candidate topic labels hierarchically according to the hierarchical relationship. When the terminal device 410 displays the candidate topic labels hierarchically, the user can select a target topic label for the first video from the hierarchical candidate topic labels, or the user can customize a target topic label for the first video.

[0417] The terminal device 410 can obtain the target topic label selected or customized by the user for the first video, and can send the target topic label corresponding to the first video to the cloud server 420. After obtaining the target topic label corresponding to the first video, the cloud server 420 can update the feature library according to the target topic label corresponding to the first video.

[0418] For example, the cloud server 420 can determine the target topic label corresponding to each image feature B, and can save each image feature B and the target topic label corresponding to each image feature B in association in the feature library.

[0419] For example, after obtaining the target topic label corresponding to the first video, the cloud server 420 can also adjust the confidence corresponding to the topic label associated with each related feature C according to the target topic label corresponding to the first video, to update the confidence corresponding to the topic label associated with each related feature C.

[0420] Please refer to FIG. 9, which shows a second schematic flowchart of a topic label determination method provided by an embodiment of the present application.

[0421] As shown in FIG. 9, when publishing a video, the user can select a first video to be published, and can edit a first title corresponding to the first video. After obtaining the first video selected by the user and the first title edited by the user, the terminal device 410 can perform feature extraction on the first title, to obtain title features.

[0422] In addition, the terminal device 410 can also perform frame extraction on the first video, to obtain M video frames, and can perform feature extraction on the M video frames respectively, to obtain image features (for example, image features A) corresponding to the video frames. Subsequently, the terminal device 410 can remove duplicates from the image features A according to the similarity between the image features A and the title features, to obtain image features B. The image features B can be all or part of the image features A.

[0423] After the terminal device 410 obtains the image features B and the title features, the terminal device 410 can splice each image feature B with the title features respectively, to obtain spliced features corresponding to each image feature B, and can send the spliced features corresponding to each image feature B to the cloud server 420.

[0424] After the cloud server 420 obtains the spliced features, the cloud server 420 can perform feature library retrieval according to each spliced feature, to determine related features (for example, related features C) similar to each spliced feature from the feature library. Subsequently, the cloud server 420 can obtain the topic labels associated with each related feature C, and can merge the topic labels associated with each related feature C, to determine candidate topic labels corresponding to the first video.

[0425] After determining the candidate topic label corresponding to the first video, the cloud server 420 can determine the hierarchical relationship between the candidate topic labels according to the common relationship of the candidate topic labels or the similarity between the candidate topic labels. Subsequently, the cloud server 420 can send the candidate topic label corresponding to the first video and the hierarchical relationship between the candidate topic labels to the terminal device 410.

[0426] After the terminal device 410 acquires the candidate topic label corresponding to the first video and the hierarchical relationship between the candidate topic labels, the terminal device 410 can display the candidate topic labels in a hierarchical manner according to the hierarchical relationship. When the terminal device 410 displays the candidate topic labels in a hierarchical manner, the user can select a target topic label for the first video from the hierarchical candidate topic labels, or the user can customize a target topic label for the first video.

[0427] The terminal device 410 can acquire the target topic label selected or customized by the user for the first video, and can send the target topic label corresponding to the first video to the cloud server 420. After acquiring the target topic label corresponding to the first video, the cloud server 420 can update the feature library according to the target topic label corresponding to the first video.

[0428] For example, the cloud server 420 can determine the target topic label corresponding to each image feature B, and can save each image feature B and the target topic label corresponding to each image feature B in association in the feature library.

[0429] For example, after acquiring the target topic label corresponding to the first video, the cloud server 420 can also adjust the confidence corresponding to the topic label associated with each related feature C according to the target topic label corresponding to the first video, to update the confidence corresponding to the topic label associated with each related feature C.

[0430] It should be noted that the terminal device 410 can include a frame extraction module, a feature extraction module, and a deduplication module. After the terminal device 410 acquires the video published by the user, the terminal device 410 can extract frames from the video through the frame extraction module to obtain video frames, and can extract features from the video through the feature extraction module to obtain image features. In addition, the terminal device 410 can also extract features from the title of the video through the feature extraction module to obtain title features. The terminal device 410 can perform deduplication processing on the image features through the deduplication module to obtain deduplicated image features.

[0431] The cloud server 420 can include a feature library module, a hierarchical construction module, and a sorting module. The cloud server 420 can obtain the image features and the title features sent by the terminal device 410. After obtaining the image features and the title features, the cloud server 420 can splice the image features and the title features through the feature library module, and can retrieve the feature library according to the spliced features to obtain relevant features. After obtaining the relevant features, the cloud server 420 can obtain the topic labels associated with each of the relevant features through the hierarchical construction module, and can merge the topic labels associated with each of the relevant features to determine candidate topic labels corresponding to the video and determine hierarchical relationships between the candidate topic labels. In addition, the cloud server 420 can sort the candidate topic labels through the sorting module. After obtaining the sorted topic titles, the cloud server 420 can send the sorted topic titles and the hierarchical relationships between the candidate topic labels to the terminal device 410, so that the terminal device 410 can display the candidate topic labels according to the hierarchical relationships.

[0432] Referring to FIG. 10, FIG. 10 shows a schematic flowchart III of a method for determining a topic label according to an embodiment of the present application.

[0433] As shown in FIG. 10, when a video is published, a user can select a first video to be published and can edit a first title corresponding to the first video. After the terminal device 410 obtains the first video selected by the user and the first title edited by the user, the terminal device 410 can extract features of the first title through a feature extraction module to obtain title features.

[0434] In addition, the terminal device 410 can also extract frames from the first video through a frame extraction module to obtain M video frames, and can extract features of the M video frames through a feature extraction module to obtain image features (for example, image features A) corresponding to each video frame. Subsequently, the terminal device 410 can remove the image features A according to the similarity between each image feature A and the title features through a de-duplication module to obtain image features B, and can splice each image feature B with the title features to obtain spliced features corresponding to each image feature B. The image features B can be all or part of the image features A. After splicing the features, the terminal device 410 can send the spliced features corresponding to each image feature B to the cloud server 420.

[0435] After the cloud server 420 acquires the spliced features, the cloud server 420 can perform feature library retrieval on the spliced features respectively according to the feature library module, to determine relevant features (for example, the relevant feature C) similar to the spliced features from the feature library. Subsequently, the cloud server 420 can acquire the topic labels associated with the relevant features C through the hierarchical construction module, and can merge the topic labels associated with the relevant features C to determine the candidate topic labels corresponding to the first video. After determining the candidate topic labels corresponding to the first video, the cloud server 420 can determine the hierarchical relationship between the candidate topic labels according to the commonality relationship of the candidate topic labels or the similarity between the candidate topic labels through the hierarchical construction module.

[0436] In addition, the cloud server 420 can also sort the candidate topic labels through the sorting module, for example, the candidate topic labels can be sorted according to the heat and / or confidence.

[0437] Subsequently, the cloud server 420 can send the sorted candidate topic labels and the hierarchical relationship between the candidate topic labels to the terminal device 410.

[0438] After the terminal device 410 acquires the candidate topic labels corresponding to the first video and the hierarchical relationship between the candidate topic labels, the terminal device 410 can display the candidate topic labels in a hierarchical manner according to the hierarchical relationship. Wherein, when the terminal device 410 displays the candidate topic labels in a hierarchical manner, the user can select a target topic label for the first video from the hierarchical candidate topic labels, or the user can customize a target topic label for the first video.

[0439] The terminal device 410 can acquire the target topic label selected or customized by the user for the first video, and can send the target topic label corresponding to the first video to the cloud server 420. After acquiring the target topic label corresponding to the first video, the cloud server 420 can update the feature library according to the target topic label corresponding to the first video.

[0440] For example, the cloud server 420 can determine the target topic labels corresponding to the image features B, and can associate and save the image features B and the target topic labels corresponding to the image features B to the feature library.

[0441] For example, after acquiring the target topic label corresponding to the first video, the cloud server 420 can also adjust the confidence corresponding to the topic labels associated with the relevant features C according to the target topic label corresponding to the first video, to update the confidence corresponding to the topic labels associated with the relevant features C.

[0442] In some embodiments, the cloud server 420 can acquire content without a topic label (which can be referred to as second content, for example), and can perform automatic completion of target topic labels for the second content without a topic label. For example, the cloud server 420 can periodically perform retrieval to determine second content without a topic label on the network. After determining the second content without a topic label, the cloud server 420 can determine candidate topic labels corresponding to the second content and hierarchical relationships between the candidate topic labels. After determining the hierarchical relationships between the candidate topic labels corresponding to the second content, the cloud server 420 can select Q candidate topic labels from the first-level candidate topic labels as target topic labels corresponding to the second content.

[0443] It should be noted that the specific content of the cloud server 420 determining the candidate topic labels corresponding to the second content and the hierarchical relationships between the candidate topic labels can refer to the related content of the cloud server 420 determining the candidate topic labels corresponding to the first content and the hierarchical relationships between the candidate topic labels, which will not be repeated here.

[0444] In one example, the cloud server 420 can select Q candidate topic labels from the first-level candidate topic labels as target topic labels corresponding to the second content according to the confidence and / or heat of each candidate topic label. For example, the cloud server 420 can determine the Q first-level candidate topic labels with the highest confidence as the target topic labels corresponding to the second content. For example, the cloud server 420 can determine the Q first-level candidate topic labels with the highest heat as the target topic labels corresponding to the second content. For example, the cloud server 420 can determine the Q first-level candidate topic labels with the highest score corresponding to the confidence and heat as the target topic labels corresponding to the second content.

[0445] In another example, the cloud server 420 can acquire the click-through rate (CTR) of the content corresponding to each candidate topic label of the first level, and can select Q candidate topic labels from the first-level candidate topic labels as target topic labels corresponding to the second content according to the click-through rate, so as to perform automatic completion of topic labels according to content with good distribution effect, which can improve the effect of automatic completion of topic labels.

[0446] For example, the cloud server 420 can determine the Q first-level candidate topic labels with the highest click-through rate of the corresponding content as the target topic labels corresponding to the second content.

[0447] In another example, the cloud server 420 can acquire the publishing time of the content corresponding to each candidate topic label in the first level, and can select Q candidate topic labels from the candidate topic labels in the first level according to the publishing time of the content, as the target topic labels corresponding to the second content, so as to improve the probability of the appearance of new topic labels and meet actual needs.

[0448] For example, the cloud server 420 can determine the Q candidate topic labels in the first level with the latest publishing time of the corresponding content as the target topic labels corresponding to the second content.

[0449] It should be noted that the cloud server 420 performs automatic completion of the topic label according to the confidence, the heat, the click rate, or the publishing time, which is only an example for illustrative explanation, and should not be understood as a limitation on the embodiments of the present application. In the embodiments of the present application, the cloud server 420 can also select Q candidate topic labels from the candidate topic labels in the first level in combination with multiple of the confidence, the heat, the click rate, or the publishing time, as the target topic labels corresponding to the second content.

[0450] It should be understood that the size of the serial number of each step in the above embodiments does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0451] Corresponding to the topic label determination method described in the above embodiments, the embodiments of the present application also provide a topic label determination device, and each module of the device can correspond to the implementation of each step of the topic label determination method.

[0452] It should be noted that the information interaction, execution process, and the like between the above devices / units are based on the same concept as the method embodiments of the present application, and the specific functions and the technical effects brought by them can be referred to the method embodiments part, which will not be repeated here.

[0453] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to complete all or part of the above described functions. Each functional unit or module in the embodiment can be integrated in one processing unit, or each unit can exist physically independently, or two or more units can be integrated in one unit, and the integrated unit can be realized in the form of hardware or in the form of software functional unit. In addition, the specific name of each functional unit or module is only for convenient distinction, and does not limit the protection scope of the present application. The specific working process of the unit or module in the system can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0454] The embodiment of the present application further provides a terminal device, which comprises at least one memory, at least one processor and a computer program stored in the at least one memory and executable on the at least one processor, and when the processor executes the computer program, the terminal device realizes the steps in any one of the method embodiments.

[0455] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and when the computer program is executed by a computer, the computer realizes the steps in any one of the method embodiments.

[0456] The embodiment of the present application provides a computer program product, which, when executed on a terminal device, enables the terminal device to realize the steps in any one of the method embodiments.

[0457] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the present application implements all or part of the processes in the above-mentioned embodiment methods, which can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer readable storage medium, and the computer program can implement the steps of each method embodiment described above when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms. The computer readable storage medium at least includes any entity or device capable of carrying the computer program code to the device / terminal equipment, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. For example, U disk, mobile hard disk, magnetic disk or optical disk, etc.

[0458] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0459] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0460] In the embodiments provided by the present application, it should be understood that the disclosed apparatus / terminal equipment and method can be implemented in other ways. For example, the above-described apparatus / terminal equipment embodiments are only schematic. For example, the division of the modules or units is only a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection between each of the units can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or in other forms.

[0461] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may also be distributed to multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiment scheme according to actual needs.

[0462] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A topic label determination method characterized by comprising: The method is applied to a terminal device, and comprises: obtaining an association relationship between a first topic label and a second topic label corresponding to first content, the first topic label and the second topic label being determined according to content features corresponding to the first content; displaying the first topic label; in response to a first operation, displaying one or more second topic labels associated with the first topic label corresponding to the first operation, the one or more second topic labels being determined according to the association relationship; in response to a second operation, determining one or more third topic labels corresponding to the second operation as target topic labels corresponding to the first content.

2. The method of claim 1, wherein, After the displaying of the one or more second topic labels associated with the first topic label corresponding to the first operation, the method further comprises: in response to a third operation, displaying one or more fourth topic labels associated with the second topic label corresponding to the third operation, the one or more fourth topic labels being determined according to the association relationship.

3. The method according to claim 1 or 2, characterized in that, The first operation comprises a selection operation, and the third topic label comprises one or more topic labels selected by the first operation.

4. The method according to any one of claims 1 to 3, characterized in that, The first operation comprises an input operation, and the third topic label comprises one or more topic labels input by the first operation.

5. The method according to any one of claims 1 to 4, characterized in that, The obtaining of the association relationship between the first topic label and the second topic label corresponding to the first content comprises: obtaining candidate topic labels corresponding to the first content, the candidate topic labels comprising the first topic label and the second topic label; determining a heat and / or a confidence corresponding to the candidate topic labels; determining the association relationship between the first topic label and the second topic label according to the heat and / or the confidence corresponding to the candidate topic labels.

6. The method of claim 5, wherein, The determining of the association relationship between the first topic label and the second topic label according to the heat and / or the confidence corresponding to the candidate topic labels comprises: for each of the candidate topic labels, determining a score corresponding to the candidate topic label according to the heat and / or the confidence corresponding to the candidate topic label; performing descending order sorting on the candidate topic labels according to the scores to obtain a first sorting result; determining the association relationship between the first topic label and the second topic label according to the first sorting result.

7. The method of claim 6, wherein, The determining of the association relationship between the first topic label and the second topic label according to the first sorting result comprises: determining a first candidate topic label in the first sorting result as a root node, the first candidate topic label being a first candidate topic label in the first sorting result; for a second candidate topic label in the first sorting result, sequentially determining a first similarity between the second candidate topic label and each of the root nodes, the second candidate topic label being any one of the first sorting result and the second candidate topic label not being the first candidate topic label; determining the second candidate topic label as a child node of the first root node when the first similarity between the second candidate topic label and the first root node is greater than or equal to a first threshold value, the first root node being any root node; determining the second candidate topic label as a new root node when the first similarity between the second candidate topic label and each of the root nodes is less than the first threshold value; determining an association relationship between the first topic label and the second topic label according to each of the root nodes and child nodes of each of the root nodes.

8. The method of claim 7, wherein, The determining an association relationship between the first topic label and the second topic label according to each of the root nodes and child nodes of each of the root nodes comprises: determining a first candidate topic label corresponding to each of the root nodes as the first topic label; determining a second candidate topic label corresponding to a child node of a second root node as a second topic label under the second root node, the second root node being any of the root nodes.

9. The method of claim 8, wherein, The determining a second candidate topic label corresponding to a child node of a second root node as a second topic label under the second root node comprises: determining a parent-child relationship between child nodes of the second root node according to the scores corresponding to the child nodes of the second root node and second similarities between the child nodes of the second root node; determining a second candidate topic label corresponding to a first parent node under the second root node as a second topic label under the second root node, and determining a third candidate topic label corresponding to a child node of the first parent node as a second topic label under the first parent node, the first parent node being any parent node under the second root node.

10. The method according to any one of claims 1 to 9, characterized in that, The obtaining an association relationship between a first topic label and a second topic label corresponding to first content comprises: obtaining the first content and sending the first content to a cloud server; obtaining, from the cloud server, an association relationship between the first topic label and the second topic label, the first topic label and the second topic label being determined by the cloud server according to content features corresponding to the first content.

11. The method according to any one of claims 1 to 9, characterized in that, The obtaining an association relationship between a first topic label and a second topic label corresponding to first content comprises: obtaining the first content and a first title corresponding to the first content, and sending the first content and the first title to a cloud server; obtaining, from the cloud server, an association relationship between the first topic label and the second topic label, the first topic label and the second topic label being determined by the cloud server according to content features corresponding to the first content and title features corresponding to the first title.

12. The method of claim 11, wherein, The sending the first content and the first title to a cloud server comprises: performing feature extraction on the first content to obtain content features corresponding to the first content; performing feature extraction on the first title to obtain title features corresponding to the first title; sending the content features and the title features to the cloud server.

13. The method of claim 12, wherein, When the first content is a video, the feature extraction on the first content to obtain the content features corresponding to the first content comprises: frame extraction on the first content to obtain first video frames corresponding to the first content; feature extraction on the first video frames to obtain the content features corresponding to the first content.

14. The method of claim 13, wherein, The feature extraction on the first video frames to obtain the content features corresponding to the first content comprises: feature extraction on the first video frames to obtain candidate features corresponding to the first content; determination of third similarities between each of the candidate features and the title features; second sorting of the candidate features in descending order according to the third similarities to obtain a second sorting result; determination of the content features corresponding to the first content according to the second sorting result.

15. The method of claim 14, wherein, The determination of the content features corresponding to the first content according to the second sorting result comprises: determination of a first candidate feature in the second sorting result as a key feature in a key feature set; for each candidate feature in the second sorting result, determination of fourth similarities between the candidate feature and each key feature in the key feature set, and determination of a candidate feature with fourth similarities smaller than a second threshold as a key feature in the key feature set; determination of the key features in the key feature set as the content features corresponding to the first content.

16. The method according to any one of claims 12 to 15, characterized in that, The sending of the content features and the title features to the cloud server comprises: for each content feature, splicing of the content feature and the title feature to obtain a spliced feature, and sending of the spliced feature to the cloud server, so that the cloud server determines relevant features from a feature library according to each spliced feature, and determines the first topic label and the second topic label corresponding to the first content according to relevant topic labels associated with the relevant features.

17. The method of any one of claims 1 to 16, wherein, After the determination of one or more third topic labels corresponding to the second operation as target topic labels corresponding to the first content in response to the second operation, the method further comprises: sending of the target topic labels to the cloud server, so that the cloud server updates the feature library in the cloud server according to the target topic labels.

18. A topic label determination method characterized by comprising: The method applied to a cloud server comprises: obtaining of content features corresponding to first content; determination of candidate topic labels corresponding to the first content according to the content features, the candidate topic labels comprising a first topic label and a second topic label; determination of an association relationship between the first topic label and the second topic label corresponding to the first content according to the candidate topic labels, the association relationship being used to determine target topic labels corresponding to the first content.

19. The method of claim 18, wherein, The method further comprises: obtaining of title features corresponding to a first title, the first title being a title of the first content; The determination of the candidate topic labels corresponding to the first content according to the content features comprises: According to the content feature and the title feature, a candidate topic label corresponding to the first content is determined.

20. The method of claim 19, wherein, The content feature corresponding to the first content is obtained by: The first content is obtained, and feature extraction is performed on the first content to obtain the content feature corresponding to the first content; The title feature of the first title is obtained by: The first title is obtained, and feature extraction is performed on the first title to obtain the title feature corresponding to the first title.

21. The method of claim 20, wherein, When the first content is a video, the content feature corresponding to the first content is obtained by: Frame extraction is performed on the first content to obtain a first video frame corresponding to the first content; Feature extraction is performed on the first video frame to obtain the content feature corresponding to the first content.

22. The method of any one of claims 18-21, wherein, After the association relationship between the first topic label and the second topic label corresponding to the first content is determined, the method further includes: The terminal device is sent the association relationship, so that the terminal device displays the topic label corresponding to the first content according to the association relationship after obtaining the association relationship; A target topic label sent by the terminal device is obtained, the target topic label including one or more topic labels selected by a user based on the topic label displayed by the terminal device and / or one or more topic labels set by the user; The feature library is updated according to the target topic label, the feature library including a plurality of preset features and topic labels associated with each preset feature.

23. The method of claim 22, wherein, The feature library is updated according to the target topic label by: For each content feature, a target topic label corresponding to the content feature is determined according to the target topic label and the related topic label associated with the related feature corresponding to the content feature, and the content feature and the target topic label corresponding to the content feature are associated and saved to the feature library.

24. The method of claim 22 or 23, wherein, The feature library is updated according to the target topic label by: According to the target topic label and the related topic label associated with each related feature corresponding to each content feature, the confidence degree corresponding to the related topic label associated with each related feature is adjusted.

25. The method of any one of claims 18-24, wherein, After the association relationship between the first topic label and the second topic label corresponding to the first content is determined, the method includes: According to the first topic label, a target topic label corresponding to the first content is determined.

26. An interactive system, characterized by The interaction system includes a terminal device and a cloud server; The cloud server is configured to obtain a content feature corresponding to first content, and determine a candidate topic label corresponding to the first content according to the content feature, the candidate topic label including a first topic label and a second topic label; The cloud server is further configured to determine an association relationship between the first topic label and the second topic label corresponding to the first content according to the candidate topic label, and send the association relationship to the terminal device; The terminal device is configured to obtain the association relationship, and display the first topic label according to the association relationship; The terminal device is further configured to, in response to a first operation, display one or more second topic labels associated with a first topic label corresponding to the first operation, the one or more second topic labels being determined according to the association relationship; The terminal device is further configured to, in response to a second operation, determine one or more third topic labels corresponding to the second operation as target topic labels corresponding to the first content.

27. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein, The processor, when executing the computer program, causes the terminal device to implement the topic label determination method according to any one of claims 1 to 17. 28.A cloud server, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein, The processor, when executing the computer program, causes the cloud server to implement the topic label determination method according to any one of claims 18 to 25.

29. A computer-readable storage medium, the computer-readable storage medium storing a computer program, characterized in that, The computer program, when executed by a computer, causes the computer to implement the topic label determination method according to any one of claims 1 to 25.

Images