Topic tag generation method, device, electronic device and storage medium

Abstract

The present disclosure relates to a method, device, electronic device and storage medium for generating topic tags, and belongs to the field of computer technology. The method includes: performing intent recognition on a live video stream, obtaining the intent type of at least one video segment in the live video stream, extracting a target video segment belonging to the target intent type from the live video stream, querying the target corpus information associated with the target video segment in the database, and generating a topic tag for the target video segment based on the target corpus information, wherein the topic tag is used to represent the subject content of the target video segment. The method provided by the embodiment of the present disclosure provides a way to automatically extract video segments, making the process of extracting video segments more intelligent, improving the convenience of the process of extracting video segments, and generating matching topic tags for video segments through the corpus information associated with the target video segment, thereby realizing the understanding of the content in the video segment and facilitating the management of the video segment.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a method, device, electronic device, and storage medium for generating a topic tag. Background Art

[0002] With the development of computer technology, live streaming has become increasingly popular, and the number of live videos produced by live streamers has increased. To facilitate the management of video streams, after the live stream ends, the live streamer usually manually segments the video into segments and assigns corresponding tags to the segments. However, this method is cumbersome. Summary of the Invention

[0003] The present disclosure provides a method, device, electronic device and storage medium for generating a topic tag, which improves the convenience of extracting video clips.

[0004] According to one aspect of an embodiment of the present disclosure, a method for generating a topic tag is provided, the method comprising:

[0005] Performing intent recognition on the live video stream to obtain an intent type of at least one video segment in the live video stream;

[0006] Extracting a target video segment belonging to a target intent type from the live video stream;

[0007] Querying the database for target corpus information associated with the target video clip;

[0008] A topic tag of the target video segment is generated based on the target corpus information, where the topic tag is used to represent the topic content of the target video segment.

[0009] In some embodiments, performing intent recognition on the live video stream to obtain the intent type of at least one video segment in the live video stream includes:

[0010] Performing voice recognition on the voice information in the live video stream to obtain text information corresponding to the voice information;

[0011] Performing intent recognition on each sentence in the text information to determine the intent type of each sentence;

[0012] According to the intent type of each sentence, the live video stream is divided into at least two video segments, each video segment only includes sentences belonging to one intent type, and the intent type corresponding to each video segment is determined.

[0013] In some embodiments, dividing the live video stream into at least two video segments according to the intent type of each sentence includes:

[0014] Determining at least two time periods based on the intent type of each sentence and the time period corresponding to each sentence, each time period including only sentences belonging to one intent type;

[0015] The live video stream is segmented according to the at least two time periods to obtain at least two video segments.

[0016] In some embodiments, performing intent recognition on each sentence in the text information and determining the intent type of each sentence includes:

[0017] Performing intent recognition on each of the sentences to determine the intent type and corresponding confidence level of each of the sentences;

[0018] Before dividing the live video stream into at least two video segments according to the intent type of each sentence, the topic tag generation method further includes:

[0019] In response to the confidence corresponding to any statement belonging to the target intent type being less than the reference confidence, the intent type of the any statement is changed to another intent type.

[0020] In some embodiments, before dividing the live video stream into at least two video segments according to the intent type of each sentence, the topic tag generation method further includes:

[0021] Determining the number of sentences having the same intention type as the target sentence based on the intention types of any adjacent reference number of sentences in the text information, where the reference number of sentences includes the target sentence, sentences before the target sentence, and sentences after the target sentence;

[0022] Determining the ratio between the determined number of sentences and the reference number as the adjustment parameter corresponding to the target sentence;

[0023] In response to the adjustment parameter being less than a reference adjustment parameter, the intent type of the target sentence is changed to another intent type.

[0024] In some embodiments, querying the database for target corpus information associated with the target video clip includes:

[0025] Determining a content type of the live video stream according to a video frame in the live video stream, where the content type is used to indicate a type of content included in the live video stream;

[0026] Query target corpus information associated with the target video clip in a database corresponding to the content type, where the object described by the target corpus information is related to the content type.

[0027] In some embodiments, determining the content type of the live video stream based on the video frames in the live video stream includes:

[0028] Performing feature extraction on the video frames in the live video stream to obtain feature vectors of the video frames;

[0029] Obtaining, based on the feature vector of the video frame, a degree of matching between the video frame and a plurality of content types;

[0030] The content type with the greatest matching degree is determined as the content type of the live video stream.

[0031] In some embodiments, obtaining the matching degree between the video frame and the multiple content types based on the feature vector of the video frame includes:

[0032] The classification model is called to perform classification processing on the feature vector to obtain a matching degree between the video frame and a plurality of content types in the classification model.

[0033] In some embodiments, before obtaining the matching degree between the video frame and multiple content types based on the feature vector of the video frame, the topic tag generation method further includes:

[0034] According to the live broadcast room information corresponding to the live broadcast room that plays the live video stream, a content type that matches the live broadcast room information is selected from multiple content types.

[0035] In some embodiments, obtaining the matching degree between the video frame and the multiple content types based on the feature vector of the video frame includes:

[0036] Performing feature extraction on live broadcast room information corresponding to the live broadcast room playing the live video stream to obtain a feature vector of the live broadcast room information;

[0037] Fusing the feature vector of the video frame with the feature vector of the live broadcast room information to obtain a fused feature vector;

[0038] The classification model is called to perform classification processing on the fused feature vector to obtain a matching degree between the video frame and a plurality of content types in the classification model.

[0039] In some embodiments, the database includes multiple pieces of corpus information and images corresponding to each piece of corpus information, and each piece of corpus information is used to describe an object included in a corresponding image;

[0040] The step of querying the database for target corpus information associated with the target video clip includes:

[0041] Performing frame extraction processing on the target video segment to obtain extracted video frames;

[0042] determining a similarity between the video frame and each image in the database;

[0043] The corpus information corresponding to the image with the maximum similarity is determined as the target corpus information.

[0044] In some embodiments, querying the database for target corpus information associated with the target video clip includes:

[0045] Obtaining text information in the target video clip;

[0046] Performing keyword extraction on the text information to obtain a plurality of keywords included in the text information;

[0047] The target corpus information matching the multiple keywords is selected from the multiple corpus information included in the database.

[0048] In some embodiments, obtaining text information of the target video segment includes:

[0049] Perform speech recognition on the speech information in the target video clip to obtain text information corresponding to the speech information.

[0050] In some embodiments, obtaining text information of the target video segment includes:

[0051] Performing frame extraction processing on the target video segment to obtain at least one video frame in the target video segment;

[0052] Text information included in the at least one video frame is extracted.

[0053] In some embodiments, generating a topic tag for the target video segment based on the target corpus information includes:

[0054] The object name included in the target corpus information is determined as the subject tag of the target video clip.

[0055] In some embodiments, after generating the topic tag of the target video segment based on the target corpus information, the topic tag generation method further includes:

[0056] The theme tag is determined to be the theme tag of the live broadcast room that plays the live video stream.

[0057] According to another aspect of the embodiments of the present disclosure, a device for generating a topic tag is provided, the device comprising:

[0058] an intent recognition unit configured to perform intent recognition on the live video stream to obtain an intent type of at least one video segment in the live video stream;

[0059] An extraction unit is configured to extract a target video segment belonging to a target intent type from the live video stream;

[0060] A query unit is configured to query the database for target corpus information associated with the target video clip;

[0061] The generating unit is configured to generate a topic tag of the target video segment based on the target corpus information, where the topic tag is used to represent the topic content of the target video segment.

[0062] In some embodiments, the intention recognition unit includes:

[0063] a speech recognition subunit, configured to perform speech recognition on the speech information in the live video stream to obtain text information corresponding to the speech information;

[0064] an intention recognition subunit, configured to perform intention recognition on each sentence in the text information and determine the intention type of each sentence;

[0065] The division subunit is configured to divide the live video stream into at least two video segments according to the intent type of each sentence, each video segment only includes sentences belonging to one intent type, and determine the intent type corresponding to each video segment.

[0066] In some embodiments, the division sub-unit is configured to determine at least two time periods based on the intent type of each statement and the time period corresponding to each statement, and each time period only includes statements belonging to one intent type; and divide the live video stream according to the at least two time periods to obtain at least two video segments.

[0067] In some embodiments, the intention recognition subunit is configured to perform intention recognition on each of the sentences and determine the intention type and corresponding confidence level of each sentence;

[0068] The subject tag generating device further includes:

[0069] The replacement unit is configured to execute, in response to the confidence corresponding to any statement belonging to the target intent type being less than the reference confidence, replacing the intent type of any statement with another intent type.

[0070] In some embodiments, the topic tag generating device further includes:

[0071] a determining unit configured to determine a number of sentences having the same intention type as the target sentence based on the intention types of any adjacent reference number of sentences in the text information, wherein the reference number of sentences includes the target sentence, sentences before the target sentence, and sentences after the target sentence;

[0072] The determining unit is further configured to determine a ratio between the determined number of sentences and the reference number as an adjustment parameter corresponding to the target sentence;

[0073] A replacement unit is configured to replace the intention type of the target sentence with another intention type in response to the adjustment parameter being less than a reference adjustment parameter.

[0074] In some embodiments, the query unit includes:

[0075] a determining subunit, configured to determine a content type of the live video stream based on video frames in the live video stream, where the content type is used to indicate a type of content contained in the live video stream;

[0076] The query subunit is configured to query the database corresponding to the content type for target corpus information associated with the target video clip, where the object described by the target corpus information is related to the content type.

[0077] In some embodiments, the determination subunit is configured to perform feature extraction on the video frames in the live video stream to obtain feature vectors of the video frames; based on the feature vectors of the video frames, obtain the matching degree between the video frames and multiple content types; and determine the content type with the greatest matching degree as the content type of the live video stream.

[0078] In some embodiments, the determination subunit is configured to execute a call to a classification model, perform classification processing on the feature vector, and obtain a matching degree between the video frame and a plurality of content types in the classification model.

[0079] In some embodiments, the topic tag generating device further includes:

[0080] The selection unit is configured to select a content type that matches the live room information from multiple content types based on the live room information corresponding to the live room that plays the live video stream.

[0081] In some embodiments, the determination subunit is configured to perform feature extraction on the live room information corresponding to the live room playing the live video stream to obtain a feature vector of the live room information; fuse the feature vector of the video frame with the feature vector of the live room information to obtain a fused feature vector; call a classification model to classify the fused feature vector to obtain a matching degree between the video frame and multiple content types in the classification model.

[0082] In some embodiments, the database includes multiple pieces of corpus information and images corresponding to each piece of corpus information, and each piece of corpus information is used to describe an object included in a corresponding image;

[0083] The query unit is configured to perform frame extraction processing on the target video segment to obtain extracted video frames;

[0084] Determine the similarity between the video frame and each image in the database; and determine the corpus information corresponding to the image corresponding to the maximum similarity as the target corpus information.

[0085] In some embodiments, the query unit is configured to obtain text information in the target video clip; perform keyword extraction on the text information to obtain multiple keywords included in the text information; and select the target corpus information that matches the multiple keywords from the multiple corpus information included in the database.

[0086] In some embodiments, the query unit is configured to perform speech recognition on the speech information in the target video segment to obtain text information corresponding to the speech information.

[0087] In some embodiments, the query unit is configured to perform frame extraction processing on the target video segment to obtain at least one video frame in the target video segment; and extract text information included in the at least one video frame.

[0088] In some embodiments, the generating unit is configured to determine the object name included in the target corpus information as the subject tag of the target video segment.

[0089] In some embodiments, the topic tag generating device further includes:

[0090] The determining unit is configured to determine the theme tag as the theme tag of the live broadcast room that plays the live video stream.

[0091] According to another aspect of the embodiments of the present disclosure, an electronic device is provided, the electronic device including:

[0092] one or more processors;

[0093] volatile or non-volatile memory for storing the one or more processor-executable instructions;

[0094] The one or more processors are configured to execute the topic tag generation method described in the first aspect.

[0095] According to another aspect of an embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the topic tag generation method described in the above aspect.

[0096] According to another aspect of the embodiments of the present disclosure, a computer program product is provided, including a computer program, which implements the topic tag generation method described in the above aspect when executed by a processor.

[0097] The topic tag generation method, device, electronic device and storage medium provided by the embodiments of the present disclosure provide a way to automatically extract video clips, extract video clips of the target intent type from a live video stream, and eliminate the need for manual extraction of video clips, making the process of extracting video clips more intelligent and improving the convenience of the process of extracting video clips. In addition, matching topic tags are generated for video clips through corpus information associated with the target video clips, thereby achieving understanding of the content in the video clips and facilitating the management of video clips.

[0098] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0100] Figure 1 It is a schematic diagram showing an implementation environment according to an exemplary embodiment.

[0101] Figure 2 The figure is a flowchart of a method for generating a topic tag according to an exemplary embodiment.

[0102] Figure 3 The figure is a flowchart of a method for generating a topic tag according to an exemplary embodiment.

[0103] Figure 4 The figure is a flowchart showing a method of generating a topic tag according to an exemplary embodiment.

[0104] Figure 5The figure is a block diagram of a device for generating a topic tag according to an exemplary embodiment.

[0105] Figure 6 The figure is a block diagram of a device for generating a topic tag according to an exemplary embodiment.

[0106] Figure 7 It is a block diagram of a terminal according to an exemplary embodiment.

[0107] Figure 8 The figure is a block diagram of a server according to an exemplary embodiment. DETAILED DESCRIPTION

[0108] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0109] It should be noted that the terms "first," "second," and the like in the specification and claims of the present disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure as detailed in the appended claims.

[0110] The terms "at least one," "a plurality," "each," and "any" as used herein include one, two, or more than two, "a plurality" includes two or more than two, "each" refers to each of the corresponding plurality, and "any" refers to any one of the plurality. For example, a plurality of statements includes three statements, and "each" refers to each of the three statements, and "any" refers to any one of the three statements, which can be the first statement, the second statement, or the third statement.

[0111] It should be noted that the user information involved in this disclosure (including but not limited to user device information, user personal information, etc.) is all information authorized by the user or fully authorized by all parties.

[0112] Figure 1 1 is a schematic diagram of an implementation environment provided according to an exemplary embodiment, the implementation environment includes: a terminal 101 and a server 102, a network connection is established between the terminal 101 and the multimedia server 102, and the terminal 101 and the multimedia server 102 can interact through the network connection.

[0113] The terminal 101 is a terminal of various types such as a mobile phone, a tablet computer, a computer, etc., and the multimedia server 102 is a server, or a server cluster composed of several servers, or a cloud computing service center.

[0114] In some embodiments, the terminal 101 is an anchor terminal, which can broadcast live video streams and upload the live video streams to the server 102. The server 102 is used to manage the live video streams of the anchor terminal, and can publish the live video streams uploaded by the anchor terminal in the live broadcast room to realize the live video broadcast of the anchor terminal. It can also capture video clips from the live video streams uploaded by the anchor terminal and generate theme tags for the video clips.

[0115] The method provided in the embodiments of the present disclosure can be applied in a variety of scenarios.

[0116] For example, in the search database construction scenario.

[0117] The terminal is installed with a live broadcast application, and the terminal uploads the live video stream to the live broadcast server through the live broadcast application. The live broadcast server uses the method provided by the embodiment of the present disclosure to extract video clips belonging to the target intent type from the live video stream, and generates a theme tag for the video clip. The theme tag and the video clip are stored in a search database corresponding to each other, so that subsequent users can search for the corresponding video clip through the theme tag.

[0118] Figure 2 This is a flowchart of a method for generating a topic tag according to an exemplary embodiment. Figure 2 , the method is applied to the server and includes the following steps:

[0119] 201. The server performs intent recognition on the live video stream to obtain an intent type of at least one video segment in the live video stream.

[0120] 202. The server extracts a target video segment belonging to the target intent type from the live video stream.

[0121] 203. The server searches the database for target corpus information associated with the target video clip.

[0122] 204. The server generates a topic tag for the target video segment based on the target corpus information. The topic tag is used to represent the topic content of the target video segment.

[0123] The method provided by the embodiment of the present disclosure provides a way to automatically extract video clips, extracting video clips of the target intent type from a live video stream without the need for manual extraction of video clips, making the process of extracting video clips more intelligent and improving the convenience of the process of extracting video clips. It also generates matching topic tags for the video clips through the corpus information associated with the target video clips, thereby realizing the understanding of the content in the video clips and facilitating the management of the video clips.

[0124] In some embodiments, performing intent recognition on a live video stream to obtain an intent type of at least one video segment in the live video stream includes:

[0125] Perform voice recognition on the voice information in the live video stream to obtain the text information corresponding to the voice information;

[0126] Perform intent recognition on each sentence in the text information and determine the intent type of each sentence;

[0127] According to the intent type of each sentence, the live video stream is divided into at least two video segments, each video segment only includes sentences belonging to one intent type, and the intent type corresponding to each video segment is determined.

[0128] In some embodiments, the live video stream is divided into at least two video segments according to the intent type of each sentence, including:

[0129] Determine at least two time periods based on the intent type of each sentence and the time period corresponding to each sentence, where each time period only includes sentences belonging to one intent type;

[0130] The live video stream is divided according to at least two time periods to obtain at least two video segments.

[0131] In some embodiments, performing intent recognition on each sentence in the text information to determine the intent type of each sentence includes:

[0132] Perform intent recognition on each sentence to determine the intent type and corresponding confidence level of each sentence;

[0133] Before dividing the live video stream into at least two video segments according to the intent type of each sentence, the method further includes:

[0134] In response to the confidence corresponding to any statement belonging to the target intent type being less than the reference confidence, the intent type of any statement is changed to another intent type.

[0135] In some embodiments, before dividing the live video stream into at least two video segments according to the intent type of each sentence, the method further includes:

[0136] Determine the number of sentences having the same intention type as the target sentence based on the intention types of any adjacent reference number of sentences in the text information, where the reference number of sentences includes the target sentence, sentences before the target sentence, and sentences after the target sentence;

[0137] The ratio between the determined number of sentences and the reference number is determined as the adjustment parameter corresponding to the target sentence;

[0138] In response to the adjustment parameter being less than the reference adjustment parameter, the intent type of the target sentence is changed to another intent type.

[0139] In some embodiments, querying the database for target corpus information associated with the target video clip includes:

[0140] Determining the content type of the live video stream based on the video frames in the live video stream, where the content type is used to indicate the type of content contained in the live video stream;

[0141] The target corpus information associated with the target video clip is searched in a database corresponding to the content type. The object described by the target corpus information is related to the content type.

[0142] In some embodiments, determining the content type of the live video stream based on video frames in the live video stream includes:

[0143] Extract features from video frames in live video streams to obtain feature vectors of video frames;

[0144] Obtaining matching degrees between the video frame and multiple content types based on a feature vector of the video frame;

[0145] The content type with the greatest matching degree is determined as the content type of the live video stream.

[0146] In some embodiments, obtaining a matching degree between a video frame and multiple content types based on a feature vector of the video frame includes:

[0147] The classification model is called to classify the feature vector to obtain the matching degree between the video frame and multiple content types in the classification model.

[0148] In some embodiments, before obtaining the matching degree between the video frame and the multiple content types based on the feature vector of the video frame, the method further includes:

[0149] According to the live broadcast room information corresponding to the live broadcast room playing the live video stream, a content type that matches the live broadcast room information is selected from multiple content types.

[0150] In some embodiments, obtaining a matching degree between a video frame and multiple content types based on a feature vector of the video frame includes:

[0151] Extract features of the live broadcast room information corresponding to the live broadcast room playing the live video stream to obtain a feature vector of the live broadcast room information;

[0152] Fusing the feature vector of the video frame with the feature vector of the live broadcast room information to obtain a fused feature vector;

[0153] The classification model is called to perform classification processing on the fused feature vector to obtain the matching degree between the video frame and multiple content types in the classification model.

[0154] In some embodiments, the database includes multiple pieces of corpus information and images corresponding to each piece of corpus information, and each piece of corpus information is used to describe an object included in a corresponding image;

[0155] Query the database for target corpus information associated with the target video clip, including:

[0156] Perform frame extraction processing on the target video clip to obtain extracted video frames;

[0157] Determine the similarity between the video frame and each image in the database;

[0158] The corpus information corresponding to the image with the maximum similarity is determined as the target corpus information.

[0159] In some embodiments, querying the database for target corpus information associated with the target video clip includes:

[0160] Get the text information in the target video clip;

[0161] Perform keyword extraction on the text information to obtain multiple keywords included in the text information;

[0162] Target corpus information matching multiple keywords is selected from multiple pieces of corpus information included in the database.

[0163] In some embodiments, obtaining text information of a target video segment includes:

[0164] Perform speech recognition on the speech information in the target video clip to obtain text information corresponding to the speech information.

[0165] In some embodiments, obtaining text information of a target video segment includes:

[0166] Performing frame extraction processing on the target video segment to obtain at least one video frame in the target video segment;

[0167] Text information included in at least one video frame is extracted.

[0168] In some embodiments, generating a topic tag for a target video segment based on target corpus information includes:

[0169] The object name included in the target corpus information is determined as the subject label of the target video clip.

[0170] In some embodiments, after generating a topic tag for a target video segment based on the target corpus information, the method further includes:

[0171] The hashtag is determined to be the hashtag of the live broadcast room that broadcasts the live video stream.

[0172] Figure 3 This is a flowchart of a method for generating a topic tag according to an exemplary embodiment. Figure 3 , the method is applied to the server and includes the following steps:

[0173] 301. The server obtains the live video stream being played in the live broadcast room.

[0174] The server is a live broadcast server or other server, the live broadcast server is used to manage the live video stream played in the live broadcast room, and the other server is used to process the live video stream. The live video stream includes video frames and voice information, the video frames are used to record the captured images, and the voice information is used to record the collected sounds.

[0175] In some embodiments, the server is a live broadcast server, and step 301 includes: the live broadcast server receives a live video stream uploaded by a host terminal. The host terminal is a terminal that logs into the live broadcast room based on the host account. After the host terminal obtains the live video stream, it uploads it to the live broadcast server, and the live broadcast server obtains the live video stream.

[0176] In a possible implementation of the above embodiment, the anchor terminal uploads the collected live video stream in real time, the live broadcast server receives the live video stream uploaded by the anchor terminal, and publishes the live video stream in the live broadcast room.

[0177] The anchor terminal uploads the captured video to the live broadcast server in real time in the form of a data stream. The live broadcast server then receives the live video stream uploaded by the anchor terminal and publishes the live video stream in the live broadcast room to achieve the live video effect of the live broadcast room, so that the audience accounts in the live broadcast room can watch the played live video stream.

[0178] In some embodiments, the server is another server, and step 301 includes: after the live broadcast server receives the live video stream uploaded by the anchor terminal, the live broadcast server sends the live video stream to the other server, and the other server receives the live video stream sent by the live broadcast server.

[0179] 302. The server performs voice recognition on the voice information in the live video stream to obtain text information corresponding to the voice information.

[0180] The text information includes multiple sentences. By performing voice recognition on the voice information included in the live video stream, the voice information is converted into text information, so that the intention type of different video clips in the live video stream can be subsequently determined through the text information.

[0181] In some embodiments, step 302 includes: the server invoking a speech recognition model to perform speech recognition on the speech information included in the live video stream to obtain text information corresponding to the speech information. The speech recognition model is used to convert the speech information into text information. The speech recognition model performs speech recognition on the speech information to ensure the accuracy of the recognized text information.

[0182] In some embodiments, step 302 includes performing speech recognition on the speech information included in the live video stream to obtain text information corresponding to the speech information and a time period corresponding to each sentence included in the text information. The time period corresponding to each sentence is used to represent the time period during which the sentence is played in the live video stream. The time period corresponding to the sentence includes the starting time point and the ending time point of the sentence in the speech information. For example, if the time period corresponding to any sentence in the text information is [2, 4], it means that the starting time point of the sentence in the live video stream is 2 seconds and the ending time point is 3 seconds.

[0183] In an embodiment of the present disclosure, a live video stream includes multiple video frames and voice information, and the multiple video frames and voice information correspond to the same time axis. By obtaining the text information corresponding to the voice information and the time period corresponding to each sentence included in the text information, the time period of each sentence in the text information in the live video stream can be determined, so that the live video stream can be subsequently segmented according to the time period corresponding to the sentence in the text information, thereby ensuring the accuracy of the segmented video segments.

[0184] 303. The server performs intent recognition on each sentence in the text information and determines the intent type of each sentence.

[0185] The intent type of each sentence is used to represent the intent indicated by the meaning of the sentence. In some embodiments, the intent type includes a target intent type or other intent types. Sentences belonging to the target intent type are sentences that explain the content, and sentences belonging to other intent types are sentences that do not explain the content, such as chat sentences. In the embodiments of the present disclosure, multiple sentences in a text message may have the same intent type or different intent types.

[0186] In some embodiments, the server invokes an intent recognition model to process the text information and determine the intent type of each sentence. The intent recognition model is a trained neural network model that is used to identify the intent type of a sentence. Determining the intent type of each sentence using the intent recognition model ensures the accuracy of the intent type.

[0187] In some embodiments, step 303 includes performing intent recognition on each sentence, determining the intent type and corresponding confidence level of each sentence. The confidence level indicates the accuracy of the intent type of each sentence. A higher confidence level indicates a higher accuracy of the intent type of the sentence, and a lower confidence level indicates a lower accuracy of the intent type of the sentence.

[0188] In a possible implementation of the above embodiment, the server calls the intent recognition model to perform intent recognition on each sentence, and obtains the intent type and corresponding confidence level of each sentence.

[0189] 304. The server divides the live video stream into at least two video segments according to the intent type of each sentence, and determines the intent type corresponding to each video segment.

[0190] Each video clip contains only sentences of one intent type, and each video clip contains at least one sentence. The intent type corresponding to each video clip is the same as the intent type of the sentence included in the video clip. Different video clips may correspond to the same or different intent types, and in at least two video clips, the intent types corresponding to two adjacent video clips are different. Two adjacent video clips refer to the same end time point of the previous video clip and the start time point of the next video clip.

[0191] For example, the live video stream is 600 seconds long and is divided into three video segments. The first segment is 120 seconds long, the second segment is 300 seconds long, and the third segment is 180 seconds long. The second segment is the target intent type, and the intent types corresponding to the first and third segments are other intent types. The end time point of the first segment is the same as the start time point of the second segment, and the end time point of the second segment is the same as the start time point of the third segment, that is, the first segment is adjacent to the second segment, and the second segment is adjacent to the third segment.

[0192] In some embodiments, step 304 includes: determining at least two time periods based on the intent type of each sentence and the time period corresponding to each sentence, and segmenting the live video stream according to the at least two time periods to obtain at least two video segments.

[0193] Each time period includes only sentences belonging to one intent type. After determining the intent types of multiple sentences and the time periods corresponding to each sentence, the time periods corresponding to multiple adjacent sentences belonging to the same intent type are combined into one time period, thereby obtaining at least two time periods. After obtaining at least two time periods, the video segments corresponding to each time period in the live video stream are determined, and the live video stream is segmented to obtain video segments corresponding to each time period, thereby obtaining at least two video segments.

[0194] Among them, after determining multiple adjacent statements belonging to the same intent type, a time period is formed by the starting time point of the time period corresponding to the first statement among the multiple statements and the ending time point of the time period corresponding to the last statement, that is, the obtained time period includes the multiple statements, and the multiple statements belong to the same intent type.

[0195] For example, the text information includes 20 sentences, the intent type of the 6th to 15th sentences is the target intent type, the intent type of the 1st to 5th sentences is other intent types, and the intent type of the 16th to 20th sentences is other intent types. Then, the starting time point of the time period corresponding to the 1st sentence and the end time point of the time period corresponding to the 5th sentence constitute a time period, the starting time point of the time period corresponding to the 6th sentence and the end time point of the time period corresponding to the 15th sentence constitute a time period, and the starting time point of the time period corresponding to the 16th sentence and the end time point of the time period corresponding to the 20th sentence constitute a time period, thereby obtaining 3 time periods.

[0196] In some embodiments, after executing step 303, the intention type and corresponding confidence of each statement are obtained. Then, before step 304, the method further includes: in response to the confidence corresponding to any statement belonging to the target intention type being less than the reference confidence, the intention type of any statement is changed to another intention type.

[0197] Among them, the reference confidence is an arbitrary value, for example, the reference confidence is 0.7 or 0.8, etc. If the confidence of the intent type of any sentence belonging to the target intent type is less than the reference confidence, it means that the accuracy of the intent type of the sentence is low, and the intent type of the sentence is replaced with another intent type to ensure that the sentence belonging to the target intent type has a high confidence, so that the accuracy of the target video clip belonging to the target intent type can be extracted later. For example, if the confidence corresponding to any sentence belonging to the target intent type is 0.6 and the reference confidence is 0.8, the intent type of the sentence is replaced from the target intent type to another intent type.

[0198] In some embodiments, before step 304, the method further includes: determining the number of sentences with the same intention type as the target sentence based on the intention types of any adjacent reference number of sentences in the text information, determining the ratio between the determined number of sentences and the reference number as the adjustment parameter corresponding to the target sentence, and in response to the adjustment parameter being less than the reference adjustment parameter, changing the intention type of the target sentence to another intention type.

[0199] The reference number is any value, such as 10 or 12, and the reference adjustment parameter is any value, such as 0.2 or 0.3. The reference number of statements includes the target statement, the statements before the target statement, and the statements after the target statement. The target statement is any statement in the reference number of statements except the first and last. For example, if the reference number is 10, the target statement is the fifth statement in the 10 statements.

[0200] After obtaining multiple sentences, the type adjustment parameter of the target sentence is determined by selecting the intention types of a reference number of adjacent sentences from the multiple sentences. If the type adjustment parameter of the target sentence is less than the reference adjustment parameter, it means that the intention type of the target sentence is inaccurate, and the intention type of the target sentence is replaced from the current intention type to another intention type to ensure the accuracy of the intention type of the target sentence after the replacement. When replacing the intention type of any sentence, the intention types of multiple sentences adjacent to the sentence are combined to ensure the accuracy of the intention type of the sentence. Moreover, according to the above method, the intention types of multiple sentences can be adjusted to ensure the accuracy of the intention type of the sentence, thereby ensuring the accuracy of the video clips subsequently segmented according to the intention type of the sentence.

[0201] Through intent recognition, the intent type determined for each sentence may be incorrect. Therefore, after determining the intent types of multiple sentences, the intent types of the multiple sentences are adjusted to ensure the accuracy of the adjusted intent types of the multiple sentences, thereby ensuring the accuracy of the subsequent video clips segmented according to the intent types of the sentences.

[0202] It should be noted that the embodiment of the present disclosure uses the voice information in the live video stream to divide at least two video segments and the intent type of each video segment from the live video stream. In another embodiment, there is no need to perform steps 302-304, and other methods can be adopted to perform intent recognition on the live video stream to obtain the intent type of at least one video segment in the live video stream.

[0203] 305. The server extracts a target video segment belonging to the target intent type from the live video stream.

[0204] After determining multiple video clips in the live video stream and the intent type corresponding to each video clip, a target video clip belonging to the target intent type is extracted from the multiple video clips according to the intent type corresponding to each video clip, that is, the target video clip only includes sentences belonging to the target intent type.

[0205] 306. The server searches the database for target corpus information associated with the target video clip.

[0206] In the disclosed embodiment, the database includes multiple pieces of corpus information, each piece of corpus information is used to describe an object, and different pieces of corpus information may describe different objects. For example, if the object is clothing, the corpus information includes the name of the clothing, the material of the clothing, the color of the clothing, the style of the clothing, the price of the clothing, etc.; or, if the object is a game, the corpus information includes the name of the game, the name of the game character, the type of game, etc. The target video clip is associated with the target corpus information, indicating that the object described by the content contained in the target video clip is the same as the object described by the target corpus information. After determining the target video clip, the target corpus information associated with the target video clip can be determined by querying the multiple pieces of corpus information in the database.

[0207] In some embodiments, the database includes multiple corpus information and images corresponding to each corpus information, and each corpus information is used to describe the object included in the corresponding image. Step 306 includes: performing frame extraction processing on the target video clip to obtain extracted video frames, determining the similarity between the video frames and each image in the database, and determining the corpus information corresponding to the image corresponding to the maximum similarity as the target corpus information.

[0208] The similarity is used to indicate the degree of similarity between a video frame and an image. The greater the similarity, the more likely the video frame and image are the same image. The extracted video frame is compared with multiple images in the database to determine the similarity between each image in the database and the video frame. The corpus information corresponding to the image with the greatest similarity is determined as the target corpus information associated with the target video segment, improving the accuracy of the target corpus information.

[0209] In one possible implementation of the above embodiment, the process of determining the similarity between a video frame and each image in a database includes obtaining feature vectors of the video frame and multiple images in the database, and determining the similarity between the video frame and each image in the database based on the feature vectors of the video frame and the multiple images. The similarity between the video frame and each image can be determined using an algorithm such as Euclidean distance or cosine similarity based on the feature vector of the video frame and the feature vector of each image.

[0210] In some embodiments, step 306 includes: obtaining text information in the target video clip, performing keyword extraction on the text information to obtain multiple keywords included in the text information, and selecting target corpus information that matches the multiple keywords from multiple corpus information included in the database.

[0211] Keywords are words describing an object in a target video clip. In some embodiments, a sequence entity extraction method is used to extract keywords from text information. By determining multiple keywords from the text information, target corpus information that matches the multiple keywords can be determined from multiple corpus information.

[0212] In a possible implementation of the above embodiment, the process of obtaining the text information of the target video segment includes the following two methods:

[0213] The first method involves performing speech recognition on the speech information in the target video clip to obtain the corresponding text information. The speech recognition model is used to convert the speech information into text information. The speech recognition model is used to perform speech recognition on the speech information to ensure the accuracy of the recognized text information.

[0214] The second method involves extracting frames from the target video segment to obtain at least one video frame, and then extracting text information from the at least one video frame. Each video frame may contain text information, and in some embodiments, optical character recognition (OCR) is used to extract keywords from the text information. By extracting text information from the image frames, it is possible to subsequently determine the associated target corpus information for the target video segment.

[0215] In a possible implementation of the above embodiment, after obtaining the text information, keyword extraction is performed on the text information to obtain multiple keywords matching the content type and the corresponding frequency of occurrence of each keyword. According to the order of the frequency of occurrence of multiple keywords from large to small, multiple corpus information are screened in turn to obtain the target corpus information.

[0216] In a possible implementation of the above embodiment, the process of extracting keywords from text information includes: extracting keywords from the text information according to multiple attribute entries matching the content type to obtain multiple keywords.

[0217] Each keyword is the attribute value of the corresponding attribute entry. For example, if the keyword is red, the corresponding attribute entry is color; or if the keyword is down, the corresponding attribute entry is material.

[0218] It should be noted that the above two methods are to determine the target corpus information associated with the target video segment from the database based on the video frames in the target video segment, or the text information in the target video segment. In another embodiment, the above two methods can be combined to ensure that the determined target corpus information is associated with the target video segment, thereby improving the accuracy of the target corpus information.

[0219] In some embodiments, the database includes multiple corpus information and images corresponding to each corpus information, and each corpus information is used to describe the object included in the corresponding image. Step 306 includes: obtaining text information in the target video clip, performing keyword extraction on the text information to obtain multiple keywords included in the text information, selecting multiple reference corpus information matching the multiple keywords from the multiple corpus information included in the database, performing frame extraction processing on the target video clip to obtain extracted video frames, determining the similarity between the video frame and the image corresponding to each reference corpus, and determining the corpus information corresponding to the image corresponding to the maximum similarity as the target corpus information.

[0220] In some embodiments, the database includes multiple corpus information and images corresponding to each corpus information, and each corpus information is used to describe the object included in the corresponding image. Step 306 includes: performing frame extraction processing on the target video clip to obtain the extracted video frame, determining the similarity between the video frame and each image in the database, determining the corpus information corresponding to the image corresponding to the similarity greater than the reference similarity as the reference corpus information, obtaining text information in the target video clip, performing keyword extraction on the text information to obtain multiple keywords included in the text information, and selecting the target corpus information that matches the multiple keywords from the multiple reference corpus information.

[0221] In some embodiments, step 306 includes the following steps 3061-3062:

[0222] 3061. Determine the content type of the live video stream based on the video frames in the live video stream.

[0223] The content type indicates the type of content contained in the live video stream. In some embodiments, the content type includes e-commerce, gaming, show, education, or pets. For example, if the live video stream is a video of the anchor user selling goods, the content type of the live video stream is e-commerce; if the live video stream contains game images and sounds, the content type of the live video stream is gaming; if the live video stream is a video of the anchor user singing, the content type of the live video stream is show.

[0224] In some embodiments, step 3061 includes: extracting features from video frames in the live video stream to obtain feature vectors of the video frames, obtaining the degree of matching between the video frames and multiple content types based on the feature vectors of the video frames, and determining the content type with the greatest matching degree as the content type of the live video stream.

[0225] The degree of match between a video frame and any content type is used to indicate the likelihood that the video frame belongs to that content type. The greater the degree of match between the video frame and the content type, the greater the likelihood that the video frame belongs to the content type. The smaller the degree of match between the video frame and the content type, the smaller the likelihood that the video frame belongs to the content type. In the disclosed embodiment, a live video stream corresponds to multiple content types. Based on the degree of match between the video frame in the live video stream and each content type, the content type with the greatest degree of match is selected as the content type of the live video stream, so that the live video stream has a high degree of match with the determined content type, thereby ensuring the accuracy of the determined content type.

[0226] In one possible implementation of the above embodiment, the process of obtaining a video frame from a live video stream includes: extracting image information to obtain any extracted video frame. For example, in a live video stream played in a live broadcast room, a video frame at any time point in the live video stream is obtained, and the content type of the live video stream is determined based on the video frame.

[0227] In a possible implementation of the above embodiment, before determining the matching degree between the video frame and multiple content types, the method further includes: selecting a content type that matches the live room information from multiple content types based on the live room information corresponding to the live room that plays the live video stream.

[0228] The live broadcast room information is information used to describe the live broadcast room. In some embodiments, the live broadcast room information includes at least one of user portrait information, live broadcast scene type, or voice type. For example, the user portrait information includes the user's gender, age, voice type, etc. The voice type includes male, female, young, mature, etc. The live broadcast scene type includes indoor type and outdoor type. If the scene included in the live video stream is shot in a room, the live broadcast scene type of the live video stream is indoor type. If the scene included in the live video stream is shot outdoors, the live broadcast scene type of the live video stream is outdoor type.

[0229] The process of determining the live scene type includes classifying video frames in a live video stream to obtain a live scene type corresponding to the live video stream. For example, feature extraction is performed on the video frame to obtain a feature vector of the video frame, a live scene classification model is invoked to process the feature vector, and a degree of match between the video frame and multiple live scene types is determined. The live scene type with the highest degree of match is determined as the live scene type corresponding to the live video stream.

[0230] The process of determining the sound type includes: obtaining voice information from the live video stream, performing voice recognition on the voice information, and determining the sound type corresponding to the live video stream. For example, a sound classification and recognition model is invoked to recognize the voice information and determine the sound type corresponding to the live video stream.

[0231] In the disclosed embodiment, there are multiple content types corresponding to the live video stream, and different content types require different live broadcast room information. Therefore, through the live broadcast room information, the content type that matches the live broadcast room information can be filtered out from the multiple content types, so that the content type of the live video stream can be determined from the multiple filtered content types, so that the content type of the live video stream matches the video frames and live broadcast room information in the live video stream, thereby ensuring the accuracy of the content type of the live video stream.

[0232] Wherein, the content type corresponds to configuration information, which is used to describe the information of the live broadcast room that matches the content type. The process of screening content types based on the live broadcast room information includes: comparing the live broadcast room information with the configuration information of multiple content types, determining the matching degree between each content type and the live broadcast room information, and selecting the content type with a matching degree greater than a reference matching degree. Wherein, the reference matching degree is an arbitrary value, such as 0.7 or 0.8.

[0233] In a possible implementation of the above embodiment, the process of determining the matching degree between the video frame and the content type includes the following three methods:

[0234] The first method is to call a classification model to classify the feature vector and obtain the matching degree between the video frame and multiple content types in the classification model. The classification model is used to determine the content type of the video frame, and the classification model includes multiple content types.

[0235] The second method involves calling multiple classification models and processing the feature vectors separately to determine the degree of match between the video frame and the content type in each classification model. Each classification model corresponds to a content type, and different classification models correspond to different content types.

[0236] The third method: extract features from the live broadcast room information corresponding to the live broadcast room that plays the live video stream to obtain a feature vector of the live broadcast room information, fuse the feature vector of the video frame with the feature vector of the live broadcast room information to obtain a fused feature vector, call the classification model, classify the fused feature vector, and obtain the matching degree between the video frame and multiple content types in the classification model.

[0237] In the disclosed embodiment, a live video stream corresponds to multiple content types. By fusing the feature vector of the video frame with the feature vector of the live broadcast room information, the obtained fused feature vector includes both the video frame information and the live broadcast room information. Through the fused feature vector, the content type of the live video stream is determined, so that the content type matches the video frame and the live broadcast room information in the live video stream, thereby ensuring the accuracy of the content type.

[0238] 3062. Query the target corpus information associated with the target video clip in the database corresponding to the content type.

[0239] The object described by the target corpus information is related to the content type. In the embodiment of the present disclosure, each content type corresponds to a database, and different content types correspond to different databases. The database corresponding to each content type includes multiple corpus information, and each corpus information describes an object related to the corresponding content type.

[0240] 307. The server generates a topic tag for the target video segment based on the target corpus information.

[0241] Thematic tags are used to represent the subject matter of the target video clip. For example, a thematic tag of "jeans" indicates that the target video clip is a video of the livestreamer selling jeans; a thematic tag of "XX game" indicates that the target video clip contains game footage. Because the target corpus information is associated with the target video clip, the thematic tags generated from the target corpus information match the target video clip, ensuring the accuracy of the thematic tags.

[0242] In some embodiments, step 307 includes determining the object name included in the target corpus information as the subject tag of the target video segment. For example, if the object name included in the target corpus information is "jeans", then "jeans" is used as the subject tag of the target video segment.

[0243] In some embodiments, after step 307, the method further includes: storing the target video segment in association with the hashtag. By storing the target video segment in association with the hashtag, the target video segment can be subsequently searched for based on the hashtag, thereby achieving accurate playback of the target video segment in the live video stream.

[0244] In some embodiments, after step 307, the method further includes determining the hashtag as the hashtag of the live broadcast room playing the live video stream. By using the hashtag as the hashtag for the live broadcast room, a solution for automatically generating hashtags for the live broadcast room is implemented. This process requires no user operation, thereby improving convenience. Furthermore, the hashtags are obtained from the video clips broadcast in the live broadcast room, so that the hashtags match the live video stream in the live broadcast room, ensuring the accuracy of the hashtags for the live broadcast room.

[0245] In some embodiments, after step 307, the method further includes: obtaining multiple target video clips and corresponding topic tags, combining video clips belonging to the same category according to the categories to which the multiple topic tags belong to obtain a video clip set, and generating a topic tag for the video clip set according to the topic tags corresponding to the video clips in the video clip set.

[0246] For example, if the subject label of one video clip is women's jeans and the subject label of another video clip is men's jeans, then these two videos are combined to obtain a video clip set whose subject label is jeans.

[0247] In one possible implementation of the above embodiment, a video clip set and its corresponding topic tags are stored in a corresponding manner. By storing the video clip set and its corresponding topic tag object, the topic tags are provided to users so that users can obtain the corresponding video clip set through the topic tags, thereby achieving accurate playback of the video clips broadcast in the live broadcast room.

[0248] The method provided by the embodiment of the present disclosure provides a way to automatically extract video clips, extracting video clips of the target intent type from a live video stream without the need for manual extraction of video clips, making the process of extracting video clips more intelligent and improving the convenience of the process of extracting video clips. It also generates matching topic tags for the video clips through the corpus information associated with the target video clips, thereby realizing the understanding of the content in the video clips and facilitating the management of the video clips.

[0249] Moreover, based on the understanding of the content in the live video stream, the voice information and video frames of the live video stream are combined in the process of extracting the target video clip from the live video stream, thereby realizing the combination of multimodal information, realizing the understanding of the content in the live video clip, and improving the accuracy of the obtained video clip and the corresponding topic label.

[0250] Figure 4 FIG. 1 is a flow chart showing a method of generating a topic tag according to an exemplary embodiment. Figure 4As shown in the figure, the live video stream being played in the live broadcast room is obtained (Living-Streaming). After obtaining the live video stream, based on the understanding of the live video stream (Live-Streaming Understanding), the target video segment belonging to the target intent type is extracted from the live video stream and the subject label is generated for the target video segment. The process includes the following four steps:

[0251] 1. Scene Routing: After acquiring the live video stream, the content type of the live video stream is determined from multiple content types through the image frames in the live video stream and the live broadcast room information including the anchor portrait information, live broadcast scene tags, etc.

[0252] 2. Info Chunking: Combining voice and text modalities, this technology uses slicing technology to segment target video segments from live video streams that match the target intent type. This technology performs intent recognition on the voice information in the live video stream to determine the intent type of the sentence in the corresponding text message. Combining the intent types of multiple sentences, it segments the live video stream to identify target video segments with the target intent type, such as product explanation videos, game content explanation videos, and singing detection videos.

[0253] 3. Content Classification: After obtaining the target video clip belonging to the target intent type, the text information in the target video clip is subjected to text recognition and sequence annotation in combination with the image modality and text modality to obtain multiple keywords that match the content type of the target video clip. Combined with the video frames in the target video clip and the obtained multiple keywords, entities are extracted from the structured database corresponding to the content type to determine the target corpus information associated with the target video clip in the structured database. For example, if the content type is e-commerce, entity extraction is performed on the structured database corresponding to the e-commerce type to determine the object described by the content contained in the target video clip, such as whether the object is clothing, cosmetics, shoes, hats, or sports goods.

[0254] 4. Topic Detection: This generates a topic tag for the target video segment based on the target corpus information associated with it. For example, the topic tag for the target video segment is "shared bicycles." This process of extracting the target video segment from the live video stream and generating topic tags combines audio, image, and text modalities, achieving multimodal information integration and ensuring the accuracy of the target video segment and its corresponding topic tag.

[0255] Figure 5is a block diagram of a device for generating topic tags according to an exemplary embodiment. Figure 5 , the topic tag generating device includes:

[0256] The intent recognition unit 501 is configured to perform intent recognition on the live video stream to obtain an intent type of at least one video segment in the live video stream;

[0257] The extraction unit 502 is configured to extract a target video segment belonging to a target intent type from the live video stream;

[0258] A query unit 503 is configured to query the database for target corpus information associated with the target video clip;

[0259] The generating unit 504 is configured to generate a topic tag of the target video segment based on the target corpus information, where the topic tag is used to represent the topic content of the target video segment.

[0260] In some embodiments, as Figure 6 As shown, the intention recognition unit 501 includes:

[0261] The speech recognition subunit 5011 is configured to perform speech recognition on the speech information in the live video stream and obtain text information corresponding to the speech information;

[0262] The intention recognition subunit 5012 is configured to perform intention recognition on each sentence in the text information and determine the intention type of each sentence;

[0263] The division subunit 5013 is configured to divide the live video stream into at least two video segments according to the intent type of each sentence, each video segment only includes sentences belonging to one intent type, and determine the intent type corresponding to each video segment.

[0264] In some embodiments, the division sub-unit 5013 is configured to determine at least two time periods based on the intent type of each statement and the time period corresponding to each statement, and each time period only includes sentences belonging to one intent type; and divide the live video stream according to at least two time periods to obtain at least two video segments.

[0265] In some embodiments, as Figure 6 As shown, the intention recognition subunit 5012 is configured to perform intention recognition on each sentence and determine the intention type and corresponding confidence level of each sentence;

[0266] The topic tag generating device also includes:

[0267] The replacement unit 505 is configured to execute, in response to the confidence corresponding to any sentence belonging to the target intent type being less than the reference confidence, replacing the intent type of any sentence with another intent type.

[0268] In some embodiments, as Figure 6 As shown, the topic tag generating device also includes:

[0269] Determining unit 506 is configured to determine the number of sentences having the same intention type as the target sentence based on the intention types of any adjacent reference number of sentences in the text information, where the reference number of sentences includes the target sentence, sentences before the target sentence, and sentences after the target sentence;

[0270] The determining unit 506 is further configured to determine the ratio between the determined number of sentences and the reference number as an adjustment parameter corresponding to the target sentence;

[0271] The replacement unit 505 is configured to replace the intention type of the target sentence with another intention type in response to the adjustment parameter being less than the reference adjustment parameter.

[0272] In some embodiments, as Figure 6 As shown, the query unit 503 includes:

[0273] The determining subunit 5031 is configured to determine the content type of the live video stream according to the video frames in the live video stream, where the content type is used to indicate the type of content contained in the live video stream;

[0274] The query subunit 5032 is configured to query the target corpus information associated with the target video clip in the database corresponding to the content type, where the object described by the target corpus information is related to the content type.

[0275] In some embodiments, the determination subunit 5031 is configured to perform feature extraction on the video frames in the live video stream to obtain a feature vector of the video frame; based on the feature vector of the video frame, obtain the matching degree between the video frame and multiple content types; and determine the content type with the greatest matching degree as the content type of the live video stream.

[0276] In some embodiments, the determination subunit 5031 is configured to execute calling of the classification model, perform classification processing on the feature vector, and obtain the matching degree between the video frame and multiple content types in the classification model.

[0277] In some embodiments, as Figure 6 As shown, the topic tag generating device also includes:

[0278] The selection unit 507 is configured to select a content type that matches the live broadcast room information from multiple content types based on the live broadcast room information corresponding to the live broadcast room that plays the live broadcast video stream.

[0279] In some embodiments, the determination subunit 5031 is configured to perform feature extraction on the live room information corresponding to the live room playing the live video stream to obtain a feature vector of the live room information; fuse the feature vector of the video frame with the feature vector of the live room information to obtain a fused feature vector; call the classification model to classify the fused feature vector to obtain the degree of matching between the video frame and multiple content types in the classification model.

[0280] In some embodiments, the database includes multiple pieces of corpus information and images corresponding to each piece of corpus information, and each piece of corpus information is used to describe an object included in a corresponding image;

[0281] The query unit 503 is configured to perform frame extraction processing on the target video segment to obtain extracted video frames; determine the similarity between the video frames and each image in the database; and determine the corpus information corresponding to the image corresponding to the maximum similarity as the target corpus information.

[0282] In some embodiments, the query unit 503 is configured to obtain text information in the target video clip; perform keyword extraction on the text information to obtain multiple keywords included in the text information; and select target corpus information that matches multiple keywords from multiple corpus information included in the database.

[0283] In some embodiments, the query unit 503 is configured to perform speech recognition on the speech information in the target video segment to obtain text information corresponding to the speech information.

[0284] In some embodiments, the query unit 503 is configured to perform frame extraction processing on the target video segment to obtain at least one video frame in the target video segment; and extract text information included in the at least one video frame.

[0285] In some embodiments, the generating unit 504 is configured to determine the object name included in the target corpus information as the subject tag of the target video segment.

[0286] In some embodiments, as Figure 6 As shown, the topic tag generating device also includes:

[0287] The determining unit 506 is configured to determine the theme tag as the theme tag of the live broadcast room that plays the live video stream.

[0288] Regarding the apparatus in the above embodiment, the specific manner in which each unit performs operations has been described in detail in the embodiment of the method, and will not be elaborated on here.

[0289] In an exemplary embodiment, an electronic device is further provided, characterized in that the electronic device includes:

[0290] one or more processors;

[0291] volatile or non-volatile memory for storing one or more processor-executable instructions;

[0292] The one or more processors are configured to execute the steps performed by the terminal or server in the above-mentioned method for generating topic tags.

[0293] In some embodiments, the electronic device is provided as a terminal. Figure 7 This is a block diagram illustrating the structure of a terminal 700 according to an exemplary embodiment. Terminal 700 may be a portable mobile terminal, such as a smartphone, tablet computer, MP3 player (Moving Picture Experts Group Audio Layer III), MP4 player (Moving Picture Experts Group Audio Layer IV), laptop computer, or desktop computer. Terminal 700 may also be referred to as user equipment, portable terminal, laptop terminal, desktop terminal, or other similar names.

[0294] The terminal 700 includes a processor 701 and a memory 702 .

[0295] The processor 701 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 701 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor 701 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 701 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 701 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0296] The memory 702 may include one or more computer-readable storage media, which may be non-transitory. The memory 702 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 702 is used to store at least one program code, which is used to be executed by the processor 701 to implement the topic tag generation method provided in the method embodiment of the present disclosure.

[0297] In some embodiments, terminal 700 may optionally include a peripheral device interface 703 and at least one peripheral device. Processor 701, memory 702, and peripheral device interface 703 may be connected via a bus or signal lines. Each peripheral device may be connected to peripheral device interface 703 via a bus, signal lines, or circuit boards. Specifically, the peripheral device may include at least one of a radio frequency circuit 704, a display screen 705, a camera assembly 706, an audio circuit 707, a positioning assembly 708, and a power supply 709.

[0298] The peripheral device interface 703 can be used to connect at least one I / O (Input / Output)-related peripheral device to the processor 701 and the memory 702. In some embodiments, the processor 701, the memory 702, and the peripheral device interface 703 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 701, the memory 702, and the peripheral device interface 703 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0299] The RF circuit 704 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 704 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 704 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. In some embodiments, the RF circuit 704 includes an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and the like. The RF circuit 704 can communicate with other terminals via at least one wireless communication protocol. Such wireless communication protocols include, but are not limited to, the World Wide Web, metropolitan area networks, intranets, various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks, and / or WiFi (Wireless Fidelity) networks. In some embodiments, the RF circuit 704 may also include circuitry related to Near Field Communication (NFC), although this disclosure is not intended to limit this.

[0300] Display screen 705 is used to display a user interface (UI). This UI may include graphics, text, icons, videos, or any combination thereof. When display screen 705 is a touchscreen display, it is also capable of collecting touch signals on or above the surface of display screen 705. These touch signals can be input as control signals to processor 701 for processing. Display screen 705 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard. In some embodiments, there can be one display screen 705, located on the front panel of terminal 700. In other embodiments, there can be at least two display screens 705, located on different surfaces of terminal 700 or in a foldable design. In other embodiments, display screen 705 can be a flexible display, located on a curved or foldable surface of terminal 700. Display screen 705 can also be configured as a non-rectangular, irregular shape, also known as a special-shaped screen. Display screen 705 can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0301] The camera assembly 706 is used to capture images or videos. In some embodiments, the camera assembly 706 includes a front camera and a rear camera. The front camera is arranged on the front panel of the terminal, and the rear camera is arranged on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera assembly 706 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.

[0302] The audio circuit 707 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals to be input into the processor 701 for processing, or input into the radio frequency circuit 704 to achieve voice communication. For the purpose of stereo sound collection or noise reduction, there can be multiple microphones, which are respectively arranged in different parts of the terminal 700. The microphone can also be an array microphone or an omnidirectional collection microphone. The speaker is used to convert the electrical signals from the processor 701 or the radio frequency circuit 704 into sound waves. The speaker can be a traditional thin film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signals into sound waves audible to humans, but also convert the electrical signals into sound waves inaudible to humans for purposes such as ranging. In some embodiments, the audio circuit 707 may also include a headphone jack.

[0303] Positioning component 708 is used to locate the current geographic location of terminal 700 to implement navigation or LBS (Location Based Service). Positioning component 708 can be a positioning component based on the US GPS (Global Positioning System), China's Beidou system, or Russia's Galileo system.

[0304] Power supply 709 is used to power various components in terminal 700. Power supply 709 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery. When power supply 709 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged via a wired line, while a wireless rechargeable battery is a battery that is charged via a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0305] In some embodiments, the terminal 700 further includes one or more sensors 710 , including but not limited to: an acceleration sensor 711 , a gyroscope sensor 712 , a pressure sensor 713 , a fingerprint sensor 714 , an optical sensor 715 , and a proximity sensor 716 .

[0306] The accelerometer 711 can detect the magnitude of acceleration along the three coordinate axes of the coordinate system established by the terminal 700. For example, the accelerometer 711 can be used to detect the components of gravity acceleration along the three coordinate axes. The processor 701 can control the display screen 705 to display the user interface in a landscape or portrait view based on the gravity acceleration signal collected by the accelerometer 711. The accelerometer 711 can also be used to collect game or user motion data.

[0307] The gyroscope sensor 712 can detect the orientation and rotation angle of the terminal 700. It can work with the accelerometer 711 to collect the user's 3D movements on the terminal 700. Based on the data collected by the gyroscope sensor 712, the processor 701 can implement the following functions: motion sensing (for example, changing the UI based on the user's tilt operation), image stabilization during shooting, game control, and inertial navigation.

[0308] The pressure sensor 713 can be set on the side frame of the terminal 700 and / or the lower layer of the display screen 705. When the pressure sensor 713 is set on the side frame of the terminal 700, it can detect the user's grip signal of the terminal 700, and the processor 701 performs left and right hand recognition or shortcut operations based on the grip signal collected by the pressure sensor 713. When the pressure sensor 713 is set on the lower layer of the display screen 705, the processor 701 controls the operable controls on the UI interface based on the user's pressure operation on the display screen 705. The operable controls include at least one of a button control, a scroll bar control, an icon control, and a menu control.

[0309] The fingerprint sensor 714 is used to collect the user's fingerprint. The processor 701 identifies the user's identity based on the fingerprint collected by the fingerprint sensor 714, or the fingerprint sensor 714 identifies the user's identity based on the collected fingerprint. When the user's identity is recognized as a trusted identity, the processor 701 authorizes the user to perform relevant sensitive operations, such as unlocking the screen, viewing encrypted information, downloading software, making payments, and changing settings. The fingerprint sensor 714 can be set on the front, back, or side of the terminal 700. When a physical button or manufacturer logo is provided on the terminal 700, the fingerprint sensor 714 can be integrated with the physical button or manufacturer logo.

[0310] The optical sensor 715 is used to detect ambient light intensity. In one embodiment, the processor 701 can control the display brightness of the display screen 705 based on the ambient light intensity detected by the optical sensor 715. Specifically, when the ambient light intensity is high, the display brightness of the display screen 705 is increased; when the ambient light intensity is low, the display brightness of the display screen 705 is decreased. In another embodiment, the processor 701 can also dynamically adjust the shooting parameters of the camera assembly 706 based on the ambient light intensity detected by the optical sensor 715.

[0311] Proximity sensor 716, also known as a distance sensor, is provided on the front panel of terminal 700. Proximity sensor 716 is used to detect the distance between the user and the front of terminal 700. In one embodiment, when proximity sensor 716 detects that the distance between the user and the front of terminal 700 is gradually decreasing, processor 701 controls display screen 705 to switch from the screen-on state to the screen-off state. When proximity sensor 716 detects that the distance between the user and the front of terminal 700 is gradually increasing, processor 701 controls display screen 705 to switch from the screen-off state to the screen-on state.

[0312] Those skilled in the art will understand that Figure 7 The structure shown in the figure does not constitute a limitation on the terminal 700, and the terminal 700 may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component arrangement.

[0313] In some embodiments, the electronic device is provided as a server. Figure 8 This is a schematic diagram of the structure of a server according to an exemplary embodiment. The server 800 may vary significantly due to different configurations or performance, and may include one or more processors (Central Processing Units, CPUs) 801 and one or more memories 802. The memories 802 store at least one program code, which is loaded and executed by the processor 801 to implement the methods provided in the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and input and output interfaces for input and output. The server may also include other components for implementing device functions, which will not be described in detail here.

[0314] In an exemplary embodiment, a non-transitory computer-readable storage medium is also provided. When the program code in the storage medium is executed by a processor of an electronic device, the electronic device is enabled to perform the steps performed by the terminal or server in the above-mentioned method for generating topic tags. In some embodiments, the storage medium can be a non-transitory computer-readable storage medium, for example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0315] In an exemplary embodiment, a computer program product is further provided, including a computer program, which implements the above-mentioned topic tag generation method when executed by a processor.

[0316] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the disclosure herein. This disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the following claims.

[0317] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for generating a topic tag, characterized in that: The topic tag generation method includes: Perform voice recognition on the voice information in the live video stream to obtain text information corresponding to the voice information; Performing intent recognition on each sentence in the text information to determine the intent type of each sentence; Dividing the live video stream into at least two video segments according to the intent type of each sentence, each video segment including only sentences belonging to one intent type, and determining the intent type corresponding to each video segment; Extracting a target video segment belonging to a target intent type from the live video stream; Querying the database for target corpus information associated with the target video clip; A topic tag of the target video segment is generated based on the target corpus information, where the topic tag is used to represent the topic content of the target video segment.

2. The method for generating a topic tag according to claim 1, wherein: The dividing the live video stream into at least two video segments according to the intent type of each sentence includes: Determining at least two time periods based on the intent type of each sentence and the time period corresponding to each sentence, each time period including only sentences belonging to one intent type; The live video stream is segmented according to the at least two time periods to obtain at least two video segments.

3. The method for generating a topic tag according to claim 1, wherein: The performing intent recognition on each sentence in the text information to determine the intent type of each sentence includes: Performing intent recognition on each of the sentences to determine the intent type and corresponding confidence level of each of the sentences; Before dividing the live video stream into at least two video segments according to the intent type of each sentence, the topic tag generation method further includes: In response to the confidence corresponding to any statement belonging to the target intent type being less than the reference confidence, the intent type of the any statement is changed to another intent type.

4. The method for generating a topic tag according to claim 1, wherein: Before dividing the live video stream into at least two video segments according to the intent type of each sentence, the topic tag generation method further includes: Determining the number of sentences having the same intention type as the target sentence based on the intention types of any adjacent reference number of sentences in the text information, where the reference number of sentences includes the target sentence, sentences before the target sentence, and sentences after the target sentence; Determining the ratio between the determined number of sentences and the reference number as the adjustment parameter corresponding to the target sentence; In response to the adjustment parameter being less than a reference adjustment parameter, the intent type of the target sentence is changed to another intent type.

5. The method for generating a topic tag according to claim 1, wherein: The step of querying the database for target corpus information associated with the target video clip includes: Determining a content type of the live video stream according to a video frame in the live video stream, where the content type is used to indicate a type of content included in the live video stream; Query target corpus information associated with the target video clip in a database corresponding to the content type, where the object described by the target corpus information is related to the content type.

6. The method for generating a topic tag according to claim 5, wherein: The determining, based on the video frames in the live video stream, the content type of the live video stream includes: Performing feature extraction on the video frames in the live video stream to obtain feature vectors of the video frames; Obtaining, based on the feature vector of the video frame, a degree of matching between the video frame and a plurality of content types; The content type with the greatest matching degree is determined as the content type of the live video stream.

7. The method for generating a topic tag according to claim 6, wherein: The obtaining, based on the feature vector of the video frame, a degree of matching between the video frame and a plurality of content types includes: The classification model is called to perform classification processing on the feature vector to obtain a matching degree between the video frame and a plurality of content types in the classification model.

8. The method for generating a topic tag according to claim 6, wherein: Before obtaining the matching degree between the video frame and the plurality of content types based on the feature vector of the video frame, the topic tag generation method further includes: According to the live broadcast room information corresponding to the live broadcast room that plays the live video stream, a content type that matches the live broadcast room information is selected from multiple content types.

9. The method for generating a topic tag according to claim 6, wherein: The obtaining, based on the feature vector of the video frame, a degree of matching between the video frame and a plurality of content types includes: Performing feature extraction on live broadcast room information corresponding to the live broadcast room playing the live video stream to obtain a feature vector of the live broadcast room information; Fusing the feature vector of the video frame with the feature vector of the live broadcast room information to obtain a fused feature vector; The classification model is called to perform classification processing on the fused feature vector to obtain a matching degree between the video frame and a plurality of content types in the classification model.

10. The method for generating a topic tag according to claim 1, wherein: The database includes multiple pieces of corpus information and images corresponding to each piece of corpus information, wherein each piece of corpus information is used to describe an object included in a corresponding image; The step of querying the database for target corpus information associated with the target video clip includes: Performing frame extraction processing on the target video segment to obtain extracted video frames; determining a similarity between the video frame and each image in the database; The corpus information corresponding to the image with the maximum similarity is determined as the target corpus information.

11. The method for generating a topic tag according to claim 1, wherein: The step of querying the database for target corpus information associated with the target video clip includes: Obtaining text information in the target video clip; Performing keyword extraction on the text information to obtain a plurality of keywords included in the text information; The target corpus information matching the multiple keywords is selected from the multiple corpus information included in the database.

12. The method for generating a topic tag according to claim 11, wherein: The obtaining of text information in the target video segment includes: Perform speech recognition on the speech information in the target video clip to obtain text information corresponding to the speech information.

13. The method for generating a topic tag according to claim 11, wherein: The obtaining of text information in the target video segment includes: Performing frame extraction processing on the target video segment to obtain at least one video frame in the target video segment; Text information included in the at least one video frame is extracted.

14. The method for generating a topic tag according to any one of claims 1 to 13, wherein: Generating the subject tag of the target video clip based on the target corpus information includes: The object name included in the target corpus information is determined as the subject tag of the target video clip.

15. The method for generating a topic tag according to any one of claims 1 to 13, characterized in that: After generating the topic tag of the target video segment based on the target corpus information, the topic tag generation method further includes: The theme tag is determined to be the theme tag of the live broadcast room that plays the live video stream.

16. A device for generating a topic tag, characterized in that: The topic tag generating device includes: Intent recognition unit, including: A speech recognition subunit is configured to perform speech recognition on the speech information in the live video stream to obtain text information corresponding to the speech information; an intention recognition subunit, configured to perform intention recognition on each sentence in the text information and determine the intention type of each sentence; a dividing subunit configured to divide the live video stream into at least two video segments according to the intent type of each sentence, each video segment including only sentences belonging to one intent type, and determine the intent type corresponding to each video segment; An extraction unit is configured to extract a target video segment belonging to a target intent type from the live video stream; A query unit is configured to query the database for target corpus information associated with the target video clip; The generating unit is configured to generate a topic tag of the target video segment based on the target corpus information, where the topic tag is used to represent the topic content of the target video segment.

17. The topic tag generating device according to claim 16, characterized in that: The division subunit is configured to determine at least two time periods based on the intention type of each sentence and the time period corresponding to each sentence, and each time period only includes sentences belonging to one intention type; according to the at least two time periods, the live video stream is divided to obtain at least two video segments.

18. The topic tag generating device according to claim 16, characterized in that: The intention recognition subunit is configured to perform intention recognition on each of the sentences and determine the intention type and corresponding confidence level of each sentence; The subject tag generating device further includes: The replacement unit is configured to execute, in response to the confidence corresponding to any statement belonging to the target intent type being less than the reference confidence, replacing the intent type of any statement with another intent type.

19. The topic tag generating device according to claim 16, characterized in that: The subject tag generating device further includes: a determining unit configured to determine a number of sentences having the same intention type as the target sentence based on the intention types of any adjacent reference number of sentences in the text information, wherein the reference number of sentences includes the target sentence, sentences before the target sentence, and sentences after the target sentence; The determining unit is further configured to determine a ratio between the determined number of sentences and the reference number as an adjustment parameter corresponding to the target sentence; A replacement unit is configured to replace the intention type of the target sentence with another intention type in response to the adjustment parameter being less than a reference adjustment parameter.

20. The topic tag generating device according to claim 16, characterized in that: The query unit includes: a determining subunit, configured to determine a content type of the live video stream based on video frames in the live video stream, where the content type is used to indicate a type of content contained in the live video stream; The query subunit is configured to query the database corresponding to the content type for target corpus information associated with the target video clip, where the object described by the target corpus information is related to the content type.

21. The topic tag generating device according to claim 20, characterized in that: The determination subunit is configured to perform feature extraction on the video frames in the live video stream to obtain feature vectors of the video frames; based on the feature vectors of the video frames, obtain the matching degree between the video frames and multiple content types; and determine the content type with the greatest matching degree as the content type of the live video stream.

22. The topic tag generating device according to claim 21, characterized in that: The determination subunit is configured to execute calling of a classification model, perform classification processing on the feature vector, and obtain a matching degree between the video frame and a plurality of content types in the classification model.

23. The topic tag generating device according to claim 21, characterized in that: The subject tag generating device further includes: The selection unit is configured to select a content type that matches the live room information from multiple content types based on the live room information corresponding to the live room that plays the live video stream.

24. The topic tag generating device according to claim 21, characterized in that: The determination subunit is configured to perform feature extraction on the live room information corresponding to the live room playing the live video stream to obtain a feature vector of the live room information; fuse the feature vector of the video frame with the feature vector of the live room information to obtain a fused feature vector; call a classification model to classify the fused feature vector to obtain a matching degree between the video frame and multiple content types in the classification model.

25. The topic tag generating device according to claim 16, characterized in that: The database includes multiple pieces of corpus information and images corresponding to each piece of corpus information, wherein each piece of corpus information is used to describe an object included in a corresponding image; The query unit is configured to perform frame extraction processing on the target video clip to obtain extracted video frames; determine the similarity between the video frames and each image in the database; and determine the corpus information corresponding to the image corresponding to the maximum similarity as the target corpus information.

26. The topic tag generating device according to claim 16, characterized in that: The query unit is configured to obtain text information in the target video clip; perform keyword extraction on the text information to obtain multiple keywords included in the text information; and select the target corpus information that matches the multiple keywords from multiple corpus information included in the database.

27. The topic tag generating device according to claim 26, characterized in that: The query unit is configured to perform voice recognition on the voice information in the target video clip to obtain text information corresponding to the voice information.

28. The topic tag generating device according to claim 26, characterized in that: The query unit is configured to perform frame extraction processing on the target video segment to obtain at least one video frame in the target video segment; and extract text information included in the at least one video frame.

29. The topic tag generating device according to any one of claims 16 to 28, characterized in that: The generating unit is configured to determine the object name included in the target corpus information as the subject tag of the target video segment.

30. The topic tag generating device according to any one of claims 16 to 28, characterized in that: The subject tag generating device further includes: The determining unit is configured to determine the theme tag as the theme tag of the live broadcast room that plays the live video stream.

31. An electronic device, characterized in that: The electronic device comprises: one or more processors; volatile or non-volatile memory for storing the one or more processor-executable instructions; The one or more processors are configured to execute the topic tag generating method according to any one of claims 1 to 15.

32. A non-transitory computer-readable storage medium, characterized in that When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the topic tag generating method according to any one of claims 1 to 15.

Images