A multimedia resource recall method, device, equipment and storage medium

Abstract

The present disclosure relates to a multimedia resource recall method and device, equipment and storage medium, and relates to the technical field of computers, which can improve the accuracy of multimedia resource search. The multimedia resource recall method comprises: acquiring a search word and a plurality of multimedia resources; determining a search word vector feature corresponding to the search word, and a plurality of multimedia resource vector features corresponding to the plurality of multimedia resources one by one; dividing the plurality of multimedia resource vector features based on the publishing accounts corresponding to the multimedia resources, to obtain a multimedia vector feature set corresponding to each publishing account; clustering the multimedia vector feature set to obtain at least one multimedia category vector feature corresponding to the publishing account; determining a target vector feature similar to the search word vector feature from the at least one multimedia category vector feature, and establishing a corresponding relationship between the target publishing account corresponding to the search word and the target vector feature.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method, apparatus, device and storage medium for recalling multimedia resources. Background Technology

[0002] With the development of computer technology, users can search for and browse multimedia resources related to search terms on their terminals. For example, a user triggers a search request from the terminal to a server, which includes the search terms. The server then returns multimedia resources related to the search terms obtained based on the search request to the terminal, which receives and displays the multimedia resources.

[0003] In related technologies, when searching for multimedia resources such as audio and video, users typically need to annotate each multimedia resource with its own text information (such as title, description, and hashtags). After parsing the terminal's search request and obtaining the search terms, the server retrieves the text information of each multimedia resource according to the search term retrieval method. If the search terms sent by the terminal match the text information of a multimedia resource, the server returns that multimedia resource to the terminal.

[0004] In the above process, since the text information is manually annotated by users, when users annotate multimedia resources with irrelevant text information, it may lead to poor accuracy of the server when searching for multimedia resources, resulting in a poor search experience for users. Summary of the Invention

[0005] This disclosure provides a multimedia resource retrieval method, apparatus, device, and storage medium, which can improve the accuracy of the server when searching for multimedia resources, thereby enriching the user's search experience.

[0006] The technical solution of this disclosure is as follows:

[0007] According to a first aspect of the present disclosure, a multimedia resource retrieval method is provided, which can be applied to an electronic device. The method may include:

[0008] Retrieve search terms and multiple multimedia resources; these multimedia resources include live streaming multimedia resources.

[0009] Determine the vector features of the search terms corresponding to the search terms, and the vector features of the multiple multimedia resources that correspond one-to-one with the multiple multimedia resources;

[0010] Based on the publishing account corresponding to the multimedia resource, the vector features of multiple multimedia resources are divided to obtain the set of multimedia vector features corresponding to each publishing account;

[0011] Clustering of the multimedia vector feature set yields at least one multimedia category vector feature corresponding to the publishing account; a multimedia category vector feature is used to represent the category of multimedia resources published by an account.

[0012] From at least one multimedia category vector feature, identify target vector features that are similar to search term vector features, and establish a correspondence between search terms and target publishing accounts corresponding to target vector features; the correspondence is used to recall multimedia resources published by target publishing accounts based on search terms.

[0013] Optionally, establish a correspondence between search terms and target publishing accounts corresponding to target vector features, including:

[0014] Obtain the set of publishing accounts corresponding to the search terms. The set of publishing accounts includes the target publishing accounts corresponding to each target vector feature.

[0015] Using search terms as index identifiers, establish a correspondence between search terms and sets of publishing accounts.

[0016] Optionally, from at least one multimedia category vector feature, target vector features similar to the search word vector features are determined, including:

[0017] Obtain the distance between the search term vector features and the vector features of each multimedia category to obtain multiple distances;

[0018] Select distances that meet preset conditions from multiple distances, and determine the multimedia category vector features corresponding to distances that meet preset conditions as target vector features similar to search word vector features; preset conditions include: distance less than a distance threshold or distance ranking before at least one preset ranking.

[0019] Optionally, determine the search term vector features corresponding to the search term, and the multiple multimedia resource vector features corresponding one-to-one with multiple multimedia resources, including:

[0020] Retrieve resource information for each of multiple multimedia resources; the resource information is used to represent the search business information for each multimedia resource based on search terms;

[0021] Based on the resource information of each multimedia resource, multiple candidate indices are determined that correspond one-to-one with the multiple multimedia resources; the candidate indices are used to construct an undirected graph structure; the undirected graph structure includes: search term nodes corresponding to search terms, and multiple multimedia resource nodes that correspond one-to-one with the multiple multimedia resources.

[0022] From a pool of candidate indices, select multiple target indices that are greater than the index threshold, and determine multiple candidate multimedia resources that correspond one-to-one with the multiple target indices.

[0023] The search term node is connected to the multiple multimedia resource nodes corresponding to the multiple candidate multimedia resources through undirected edges to obtain an undirected graph structure;

[0024] Graph neural networks are trained on nodes in an undirected graph structure to obtain search word vector features and multiple multimedia resource vector features.

[0025] Optionally, the resource information of a multimedia resource includes: number of clicks, number of search impressions, number of times the playback duration exceeds a first threshold, number of plays, and playback duration; based on the resource information of each multimedia resource, multiple candidate indices corresponding one-to-one with multiple multimedia resources are determined, including:

[0026] For each multimedia resource's resource information, the following operations are performed to obtain multiple candidate indices corresponding one-to-one with the multiple multimedia resources:

[0027] The ratio between the number of clicks and the number of search impressions is determined as the click-through rate; the ratio between the number of times the playback duration exceeds the first threshold and the number of playbacks is determined as the playback rate; based on the click-through rate, playback rate, and playback duration, candidate indices corresponding to a multimedia resource are determined.

[0028] Optionally, a graph neural network is trained on the nodes in the undirected graph structure to obtain search word vector features and multiple multimedia resource vector features, including:

[0029] Find the first neighboring node that is connected to the search term node via an undirected edge;

[0030] Find the second neighboring node that is connected to each of the multiple multimedia resource nodes via an undirected edge;

[0031] Based on the search term node, the first neighboring node, each multimedia resource node and the second neighboring node, a graph neural network is trained on the nodes in the undirected graph structure until the undirected graph structure is trained to a convergent state, so as to obtain the search term vector features and multiple multimedia resource vector features.

[0032] According to a second aspect of the present disclosure, a multimedia resource retrieval method is provided, which can be applied to an electronic device. The method may include:

[0033] Receive search requests that include the target search term;

[0034] In response to a search request, at least one publishing account corresponding to the target search term is determined from the pre-established correspondence between search terms and publishing accounts. When establishing the correspondence, target vector features similar to the search term vector features of the search term are obtained from at least one multimedia category vector feature corresponding to the publishing account, and a correspondence between the search term and the target publishing account corresponding to the target vector feature is established.

[0035] Send multimedia resources published by at least one publishing account.

[0036] According to a third aspect of the present disclosure, a multimedia resource retrieval device is provided, which can be applied to an electronic device, including: an acquisition unit and a processing unit;

[0037] The acquisition unit is used to acquire search terms and multiple multimedia resources; these multimedia resources include live streaming multimedia resources.

[0038] The processing unit is used to determine the search term vector features corresponding to the search term, and the multiple multimedia resource vector features corresponding one-to-one with multiple multimedia resources;

[0039] The processing unit is also used to divide multiple multimedia resource vector features based on the publishing account corresponding to the multimedia resource, and obtain a set of multimedia vector features corresponding to each publishing account.

[0040] The processing unit is also used to cluster the multimedia vector feature set to obtain at least one multimedia category vector feature corresponding to the publishing account; a multimedia category vector feature is used to represent the category of multimedia resources published by an account;

[0041] The processing unit is also used to determine target vector features similar to search term vector features from at least one multimedia category vector feature, and to establish a correspondence between the search term and the target publishing account corresponding to the target vector feature; the correspondence is used to recall multimedia resources published by the target publishing account based on the search term.

[0042] Optional, processing unit, specifically used for:

[0043] Obtain the set of publishing accounts corresponding to the search terms. The set of publishing accounts includes the target publishing accounts corresponding to each target vector feature.

[0044] Using search terms as index identifiers, establish a correspondence between search terms and sets of publishing accounts.

[0045] Optional, processing unit, specifically used for:

[0046] Obtain the distance between the search term vector features and the vector features of each multimedia category to obtain multiple distances;

[0047] Select distances that meet preset conditions from multiple distances, and determine the multimedia category vector features corresponding to distances that meet preset conditions as target vector features similar to search word vector features; preset conditions include: distance less than a distance threshold or distance ranking before at least one preset ranking.

[0048] Optional, processing unit, specifically used for:

[0049] Retrieve resource information for each of multiple multimedia resources; the resource information is used to represent the search business information for each multimedia resource based on search terms;

[0050] Based on the resource information of each multimedia resource, multiple candidate indices are determined that correspond one-to-one with the multiple multimedia resources; the candidate indices are used to construct an undirected graph structure; the undirected graph structure includes: search term nodes corresponding to search terms, and multiple multimedia resource nodes that correspond one-to-one with the multiple multimedia resources.

[0051] From a pool of candidate indices, select multiple target indices that are greater than the index threshold, and determine multiple candidate multimedia resources that correspond one-to-one with the multiple target indices.

[0052] The search term node is connected to the multiple multimedia resource nodes corresponding to the multiple candidate multimedia resources through undirected edges to obtain an undirected graph structure;

[0053] Graph neural networks are trained on nodes in an undirected graph structure to obtain search word vector features and multiple multimedia resource vector features.

[0054] Optionally, the resource information of a multimedia resource includes: number of clicks, number of search impressions, number of times the playback duration exceeds a first threshold, number of plays, and playback duration; the processing unit is specifically used for:

[0055] For each multimedia resource's resource information, the following operations are performed to obtain multiple candidate indices corresponding one-to-one with the multiple multimedia resources:

[0056] The ratio between the number of clicks and the number of search impressions is determined as the click-through rate; the ratio between the number of times the playback duration exceeds the first threshold and the number of playbacks is determined as the playback rate; based on the click-through rate, playback rate, and playback duration, candidate indices corresponding to a multimedia resource are determined.

[0057] Optional, processing unit, specifically used for:

[0058] Find the first neighboring node that is connected to the search term node via an undirected edge;

[0059] Find the second neighboring node that is connected to each of the multiple multimedia resource nodes via an undirected edge;

[0060] Based on the search term node, the first neighboring node, each multimedia resource node and the second neighboring node, a graph neural network is trained on the nodes in the undirected graph structure until the undirected graph structure is trained to a convergent state, so as to obtain the search term vector features and multiple multimedia resource vector features.

[0061] According to a fourth aspect of the present disclosure, a multimedia resource retrieval device is provided, which can be applied to an electronic device, including: a receiving unit, a processing unit, and a sending unit;

[0062] A receiving unit is used to receive a search request that includes the target search term;

[0063] The processing unit is configured to respond to a search request by determining at least one publishing account corresponding to the target search term from a pre-established correspondence between search terms and publishing accounts. When establishing the correspondence, the unit obtains target vector features similar to the search term vector features from at least one multimedia category vector feature corresponding to the publishing account, and establishes a correspondence between the search term and the target publishing account corresponding to the target vector feature.

[0064] A sending unit is used to send multimedia resources published by at least one publishing account.

[0065] According to a fifth aspect of the present disclosure, an electronic device is provided, which may include: a processor and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement any of the optional multimedia resource retrieval methods of the first or second aspect described above.

[0066] According to a sixth aspect of the present disclosure, a computer-readable storage medium is provided, on which instructions are stored, such that when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform any of the optional multimedia resource retrieval methods of the first or second aspect described above.

[0067] According to a seventh aspect of the present disclosure, a computer program product is provided, the computer program product including computer instructions that, when executed on an electronic device, cause the electronic device to perform a multimedia resource recall method as described in any optional implementation of the first or second aspect.

[0068] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0069] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0070] Based on any of the above aspects, in this disclosure, after obtaining the search term and multiple multimedia resources, the search term vector features corresponding to the search term and the multiple multimedia resource vector features corresponding one-to-one with the multiple multimedia resources can be determined. The multiple multimedia resource vector features are then divided based on the publishing accounts corresponding to the multimedia resources, resulting in a set of multimedia vector features corresponding to each publishing account. Next, the multimedia vector feature sets can be clustered to obtain at least one multimedia category vector feature corresponding to each publishing account (a multimedia category vector feature represents the category of multimedia resources published by an account). From the at least one multimedia category vector feature, target vector features similar to the search term vector features are determined, and a correspondence is established between the search term and the target publishing account corresponding to the target vector feature. Since the correspondence is used to recall multimedia resources published by the target publishing account based on the search term, this disclosure can accurately recall multimedia resources published by the target publishing account based on the search term, solving the technical problem of poor accuracy in multimedia resource searches in the prior art, improving the accuracy of multimedia resource searches, and thus enriching the user's search experience. Attached Figure Description

[0071] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0072] Figure 1 A flowchart illustrating a multimedia resource retrieval method provided in an embodiment of this disclosure is shown.

[0073] Figure 2A A flowchart illustrating another multimedia resource retrieval method provided in this disclosure is shown.

[0074] Figure 2B A flowchart illustrating another multimedia resource retrieval method provided in this disclosure is shown.

[0075] Figure 3 A flowchart illustrating another multimedia resource retrieval method provided in this disclosure is shown.

[0076] Figure 4 A flowchart illustrating another multimedia resource retrieval method provided in this disclosure is shown.

[0077] Figure 5 A flowchart illustrating another multimedia resource retrieval method provided in this disclosure is shown.

[0078] Figure 6 A flowchart illustrating another multimedia resource retrieval method provided in this disclosure is shown.

[0079] Figure 7A This diagram illustrates the structure of a multimedia resource retrieval device according to an embodiment of the present disclosure.

[0080] Figure 7B This illustration shows a structural schematic diagram of yet another multimedia resource recall device provided in an embodiment of the present disclosure;

[0081] Figure 8 A schematic diagram of the structure of a terminal provided in an embodiment of this disclosure is shown;

[0082] Figure 9 A schematic diagram of the structure of a server provided in an embodiment of this disclosure is shown. Detailed Implementation

[0083] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0084] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0085] It should also be understood that the term "comprising" indicates the presence of the described feature, whole, step, operation, element and / or component, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements and / or components.

[0086] The data disclosed herein may be data authorized by the user or fully authorized by all parties.

[0087] As described in the background section, when searching for multimedia resources such as audio and video, users typically need to annotate each resource with its own text information (such as title, description, and hashtags). After parsing the terminal's search request and obtaining the search terms, the server retrieves the text information of each multimedia resource according to the search term retrieval method. If the search terms sent by the terminal match the text information of a multimedia resource, the server returns that multimedia resource to the terminal.

[0088] In the above process, since the text information is manually annotated by users, when users annotate multimedia resources with irrelevant text information, it may lead to poor accuracy of the server when searching for multimedia resources, resulting in a poor search experience for users.

[0089] Based on this, embodiments of this disclosure provide a multimedia resource retrieval method. After obtaining search terms and multiple multimedia resources, it can determine the search term vector features corresponding to the search terms and the multiple multimedia resource vector features that correspond one-to-one with the multiple multimedia resources. The multiple multimedia resource vector features are then divided based on the publishing accounts corresponding to the multimedia resources, resulting in a set of multimedia vector features corresponding to each publishing account. Next, the multimedia vector feature sets can be clustered to obtain at least one multimedia category vector feature corresponding to each publishing account (a multimedia category vector feature represents the category of multimedia resources published by an account). From the at least one multimedia category vector feature, target vector features similar to the search term vector features are determined, and a correspondence is established between the search terms and the target publishing accounts corresponding to the target vector features. Since the correspondence is used to retrieve multimedia resources published by target publishing accounts based on search terms, this disclosure can accurately retrieve multimedia resources published by target publishing accounts based on search terms, solving the technical problem of poor accuracy in multimedia resource searches in existing technologies, improving the accuracy of multimedia resource searches, and thus enriching the user's search experience.

[0090] The following provides an exemplary description of the multimedia resource retrieval method provided in the embodiments of this disclosure:

[0091] The multimedia resource retrieval method disclosed herein can be applied to electronic devices.

[0092] In some embodiments, the electronic device may be a server, a terminal, or other electronic devices used for multimedia resource retrieval, and this disclosure does not limit the scope of the device.

[0093] The server can be a single server or a server cluster consisting of multiple servers. In some embodiments, the server cluster can also be a distributed cluster. This disclosure does not limit the specific implementation of the server.

[0094] The terminal can be a mobile phone, tablet computer, desktop computer, laptop computer, handheld computer, notebook computer, ultra-mobile personal computer (UMPC), netbook, as well as cellular phone, personal digital assistant (PDA), augmented reality (AR) / virtual reality (VR) device, etc., that can install and use content community applications (such as Kuaishou). This disclosure does not impose any special restrictions on the specific form of the electronic device. It can interact with the user through one or more methods such as keyboard, touchpad, touch screen, remote control, voice interaction, or handwriting device.

[0095] The multimedia resource retrieval method provided in this application embodiment will be described in detail below with reference to the accompanying drawings.

[0096] like Figure 1 As shown, when the multimedia resource retrieval method is applied to an electronic device, the multimedia resource retrieval method may include:

[0097] S101, Electronic devices acquire search terms and multiple multimedia resources.

[0098] Among them, multiple multimedia resources include live streaming multimedia resources.

[0099] Specifically, when searching for multimedia resources of the live streaming type, users usually need to annotate each multimedia resource with its own text information (such as title, description, hashtags, etc.). Therefore, after obtaining the search terms, the electronic device can retrieve the text information of each multimedia resource according to the search term retrieval method, and identify the multiple multimedia resources corresponding to multiple text information that the search term can match as multiple multimedia resources related to the search term.

[0100] For example, if the search term is "basketball," then the electronic device, in addition to obtaining the search term "basketball," can also obtain multiple basketball-related live video resources based on the search term "basketball."

[0101] Among them, live streaming multimedia resources can be currently live streaming indicator resources, or multimedia resources formed through live streaming during a historical period (i.e., live streaming replay multimedia resources).

[0102] S102. The electronic device determines the vector features of the search term corresponding to the search term, and the vector features of multiple multimedia resources that correspond one-to-one with multiple multimedia resources.

[0103] Specifically, after acquiring the search term and multiple multimedia resources, the electronic device needs to determine the multimedia resources most relevant to the search term. Therefore, the electronic device can determine the search term vector features corresponding to the search term and the multiple multimedia resource vector features that correspond one-to-one with the multiple multimedia resources. This enables the subsequent determination of target vector features similar to the search term vector features based on the search term vector features and the multiple multimedia resource vector features that correspond one-to-one with the multiple multimedia resources.

[0104] Optionally, the electronic device determines the search term vector features corresponding to the search term and the multiple multimedia resource vector features corresponding to multiple multimedia resources one-to-one through a vector feature extraction algorithm; it can also determine the search term vector features corresponding to the search term and the multiple multimedia resource vector features corresponding to multiple multimedia resources one-to-one through an undirected graph structure and a graph neural network; or it can determine the search term vector features corresponding to the search term and the multiple multimedia resource vector features corresponding to multiple multimedia resources one-to-one through other methods, which are not limited in this disclosure.

[0105] S103. The electronic device divides the vector features of multiple multimedia resources based on the publishing accounts corresponding to the multimedia resources, and obtains the set of multimedia vector features corresponding to each publishing account.

[0106] Specifically, after determining the search term vector features corresponding to the search term and the multiple multimedia resource vector features corresponding one-to-one with multiple multimedia resources, the electronic device can divide the multiple multimedia resource vector features based on the publishing account corresponding to the multimedia resource to obtain the multimedia vector feature set corresponding to each publishing account.

[0107] For example, publishing account A publishes 3 multimedia resources, and publishing account B publishes 5 multimedia resources. In this case, the multimedia vector features of the 3 multimedia resources published by publishing account A can be determined as the multimedia vector feature set corresponding to publishing account A. The multimedia vector features of the 5 multimedia resources published by publishing account B can be determined as the multimedia vector feature set corresponding to publishing account B.

[0108] S104. The electronic device clusters the multimedia vector feature set to obtain at least one multimedia category vector feature corresponding to the publishing account.

[0109] Specifically, after dividing the vector features of multiple multimedia resources based on the publishing accounts corresponding to the multimedia resources, and obtaining the set of multimedia vector features corresponding to each publishing account, the set of multimedia vector features can be clustered to obtain at least one multimedia category vector feature corresponding to the publishing account.

[0110] One of the features is a multimedia category vector, which is used to represent the category of multimedia resources published by an account.

[0111] Optionally, when clustering multimedia vector feature sets, electronic devices can cluster the multimedia vector feature sets according to a preset number of categories.

[0112] Optionally, when the electronic device clusters the multimedia vector feature set, it may use the K-Nearest Neighbor (KNN) classification algorithm to cluster the multimedia vector feature set; it may also use the k-means clustering algorithm to cluster the multimedia vector feature set; or it may use a clustering algorithm to cluster the multimedia vector feature set. This disclosure does not limit the choice of which method to use.

[0113] For example, an electronic device obtains a set of multimedia vector features corresponding to 30 multimedia resources published by account A. Among them, 15 are multimedia vector features corresponding to "sports" multimedia resources, and 15 are multimedia vector features corresponding to "music" multimedia resources. In this case, the electronic device clusters the set of multimedia vector features corresponding to the 30 multimedia resources based on the K-nearest neighbor classification algorithm, obtaining two multimedia category vector features corresponding to account A: multimedia category vector features for "sports" and multimedia category vector features for "music".

[0114] S105. The electronic device determines target vector features similar to search term vector features from at least one multimedia category vector feature, and establishes a correspondence between the search term and the target publishing account corresponding to the target vector feature.

[0115] Specifically, after obtaining at least one multimedia category vector feature corresponding to the publishing account, the electronic device can determine the target vector feature that is similar to the search term vector feature from the at least one multimedia category vector feature, and establish the correspondence between the search term and the target publishing account corresponding to the target vector feature.

[0116] The correspondence is used to recall multimedia resources published by target publishing accounts based on search terms.

[0117] Based on the above example, the preset search term is "basketball". The electronic device clusters the multimedia vector feature sets corresponding to 30 multimedia resources, obtaining two multimedia category vector features corresponding to publishing account A: the multimedia category vector features for "sports" and the multimedia category vector features for "music". The electronic device can then determine that the search term vector feature for "basketball" is similar to the multimedia category vector feature for "sports" among the two multimedia category vector features mentioned above. Specifically, the multimedia category vector features for "sports" include the multimedia category vector features corresponding to the multimedia resources for "basketball games".

[0118] In this scenario, the electronic device can identify target vector features similar to the search term vector features of the search term "basketball" from the multimedia category vector features of "sports". The target vector features are the multimedia category vector features corresponding to the multimedia resources of "basketball game". The device can then establish a correspondence between the target publishing accounts corresponding to the multimedia category vector features of the multimedia resources of "basketball game" and the search term "basketball".

[0119] Subsequently, when users search for the term "basketball," the multimedia resources published by the account that posted "basketball games" can be recalled.

[0120] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S101-S105, after obtaining the search term and multiple multimedia resources, the search term vector features corresponding to the search term and the multiple multimedia resource vector features corresponding one-to-one with the multiple multimedia resources can be determined. Based on the publishing accounts corresponding to the multimedia resources, the multiple multimedia resource vector features are divided to obtain a set of multimedia vector features corresponding to each publishing account. Next, the multimedia vector feature sets can be clustered to obtain at least one multimedia category vector feature corresponding to the publishing account (a multimedia category vector feature is used to represent the category of multimedia resources published by an account). From the at least one multimedia category vector feature, target vector features similar to the search term vector features are determined, and a correspondence between the search term and the target publishing account corresponding to the target vector feature is established. Since the correspondence is used to recall multimedia resources published by the target publishing account based on the search term, this disclosure can accurately recall multimedia resources published by the target publishing account based on the search term, solving the technical problem of poor accuracy in multimedia resource searches in the prior art, improving the accuracy of multimedia resource searches, and thus enriching the user's search experience.

[0121] In one embodiment, such as Figure 2A As shown, in S105 above, the method for establishing the correspondence between search terms and target publishing accounts corresponding to target vector features in electronic devices specifically includes:

[0122] S201. Obtain the set of publishing accounts corresponding to the search terms for electronic devices.

[0123] Specifically, when an electronic device establishes a correspondence between search terms and target publishing accounts corresponding to target vector features, it can obtain a set of publishing accounts corresponding to the search terms.

[0124] The set of publishing accounts includes the target publishing accounts corresponding to each target vector feature.

[0125] S202. Electronic devices establish a correspondence between search terms and a set of publishing accounts, using search terms as index identifiers.

[0126] Specifically, after obtaining the set of publishing accounts corresponding to the search terms, the electronic device establishes a correspondence between the search terms and the set of publishing accounts, using the search terms as index identifiers.

[0127] Optionally, this correspondence can be an inverted index.

[0128] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S201-S202, when the electronic device establishes the correspondence between search terms and target publishing accounts corresponding to target vector features, it can obtain the set of publishing accounts corresponding to the search terms, and establish the correspondence between search terms and the set of publishing accounts using the search terms as index identifiers. This provides a specific implementation method for establishing the correspondence between search terms and target publishing accounts corresponding to target vector features. Since the correspondence is used to recall multimedia resources published by target publishing accounts based on search terms, the electronic device can accurately recall multimedia resources published by target publishing accounts based on search terms, solving the technical problem of poor accuracy in multimedia resource searches in the prior art, improving the accuracy of multimedia resource searches, and thus enriching the user's search experience.

[0129] In one embodiment, such as Figure 2B As shown, in S105 above, the method by which the electronic device determines target vector features similar to search word vector features from at least one multimedia category vector feature specifically includes:

[0130] S203. The electronic device obtains the distance between the search term vector features and the vector features of each multimedia category to obtain multiple distances.

[0131] Specifically, when an electronic device determines a target vector feature that is similar to a search word vector feature from at least one multimedia category vector feature, it can obtain the distance between the search word vector feature and each multimedia category vector feature to obtain multiple distances.

[0132] Optionally, when an electronic device obtains the distance between the search term vector features and each multimedia category vector feature, it can use a distance calculation method to obtain the distance between the search term vector features and each multimedia category vector feature.

[0133] Alternatively, the distance calculation method can be the Cosine distance method, the Euclidean distance method, etc.

[0134] S204. The electronic device selects a distance that meets preset conditions from multiple distances, and determines the multimedia category vector feature corresponding to the distance that meets the preset conditions as the target vector feature similar to the search word vector feature.

[0135] Specifically, after obtaining the distance between the search term vector features and each multimedia category vector feature to obtain multiple distances, the electronic device selects the distance that meets the preset conditions from the multiple distances, and determines the multimedia category vector feature corresponding to the distance that meets the preset conditions as the target vector feature similar to the search term vector features.

[0136] The preset conditions include: the distance is less than the distance threshold or the distance is sorted before at least one preset sort.

[0137] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S203-S204, when the electronic device determines a target vector feature similar to the search term vector feature from at least one multimedia category vector feature, it can obtain the distance between the search term vector feature and each multimedia category vector feature to obtain multiple distances. It then selects the distance that meets a preset condition from these multiple distances and determines the multimedia category vector feature corresponding to the distance that meets the preset condition as the target vector feature similar to the search term vector feature. This provides a specific implementation method for determining target vector features similar to the search term vector feature, enabling the subsequent establishment of a correspondence between the search term and the target publishing account corresponding to the target vector feature. Since the correspondence is used to recall multimedia resources published by the target publishing account based on the search term, the electronic device can accurately recall multimedia resources published by the target publishing account based on the search term, solving the technical problem of poor accuracy in multimedia resource searching in the prior art, improving the accuracy of multimedia resource searching, and thus enriching the user's search experience.

[0138] In one embodiment, combined with Figure 1 ,like Figure 3 As shown, in S102 above, the method for the electronic device to determine the search term vector features corresponding to the search term and the vector features of multiple multimedia resources corresponding one-to-one with multiple multimedia resources specifically includes:

[0139] S301. The electronic device acquires resource information for each of the multiple multimedia resources.

[0140] Specifically, when an electronic device determines the vector features of the search term corresponding to the search term, and the vector features of multiple multimedia resources that correspond one-to-one with multiple multimedia resources, it can obtain the resource information of each multimedia resource among the multiple multimedia resources.

[0141] The resource information is used to represent the search business information for each multimedia resource based on search terms. The resource information of a multimedia resource includes: number of clicks, number of search impressions, number of times the playback duration exceeds a first threshold, number of plays, and playback duration.

[0142] For example, when searching for the term "basketball," multiple multimedia resources corresponding to the term "basketball" can be retrieved. In this case, the electronic device can retrieve resource information for each multimedia resource, such as the number of clicks, the number of search impressions, the number of times the playback duration exceeds a first threshold, the number of plays, and the playback duration.

[0143] S302. The electronic device determines multiple candidate indices that correspond one-to-one with each multimedia resource based on the resource information of each multimedia resource.

[0144] Specifically, after acquiring the resource information of each multimedia resource among multiple multimedia resources, the electronic device can determine multiple candidate indices that correspond one-to-one with the multiple multimedia resources based on the resource information of each multimedia resource.

[0145] The candidate index is used to represent the importance of each multimedia resource. By determining the candidate index, multimedia resources with higher importance can be selected to construct an undirected graph structure, thereby determining the vector characteristics of the multimedia resources.

[0146] S303. The electronic device selects multiple target indices that are greater than the index threshold from multiple candidate indices, and determines multiple candidate multimedia resources that correspond one-to-one with the multiple target indices.

[0147] Specifically, after determining multiple candidate indices corresponding one-to-one with each multimedia resource based on the resource information of each multimedia resource, the electronic device can select multiple target indices that are greater than the index threshold from the multiple candidate indices, and determine multiple candidate multimedia resources corresponding one-to-one with the multiple target indices.

[0148] S304. The electronic device connects the search term node with multiple multimedia resource nodes corresponding to multiple candidate multimedia resources through undirected edges to obtain an undirected graph structure.

[0149] Specifically, after selecting multiple target indices that are greater than the index threshold from multiple candidate indices, and determining multiple candidate multimedia resources that correspond one-to-one with the multiple target indices, the electronic device can connect the search term node with the multiple multimedia resource nodes corresponding to the multiple candidate multimedia resources through undirected edges to obtain an undirected graph structure.

[0150] Undirected graphs are a type of data structure where edges are undirected, meaning that two connected vertices can reach each other. By constructing undirected graph structures, search word vector features and multiple multimedia resource vector features can be trained.

[0151] In one embodiment, the undirected graph structure includes: a search term node corresponding to a search term, and multiple multimedia resource nodes corresponding one-to-one with multiple multimedia resources. Undirected edges exist between the search term nodes and some of the multimedia resource nodes. These undirected edges can be determined based on the resource information of the multimedia resources corresponding to the multimedia resource nodes.

[0152] S305. The electronic device trains a graph neural network on the nodes in the undirected graph structure to obtain search word vector features and multiple multimedia resource vector features.

[0153] Specifically, after obtaining the undirected graph structure, electronic devices can train a graph neural network (GNN) on the nodes in the undirected graph structure to obtain search word vector features and multiple multimedia resource vector features.

[0154] Graph neural networks are a type of neural network that operates directly on graph structures. In this disclosure, an electronic device can utilize a graph neural network to train nodes in an undirected graph structure to obtain search word vector features and multiple multimedia resource vector features.

[0155] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S301-S305, when the electronic device determines the search term vector features corresponding to the search term and the multiple multimedia resource vector features corresponding one-to-one with the multiple multimedia resources, it can obtain the resource information of each multimedia resource among the multiple multimedia resources, and determine multiple candidate indices corresponding one-to-one with the multiple multimedia resources based on the resource information of each multimedia resource. Next, the electronic device selects multiple target indices greater than the index threshold from the multiple candidate indices, determines multiple candidate multimedia resources corresponding one-to-one with the multiple target indices, and connects the search term node with the multiple multimedia resource nodes corresponding to the multiple candidate multimedia resources through undirected edges to obtain an undirected graph structure. Subsequently, the electronic device performs graph neural network training on the nodes in the undirected graph structure to obtain the search term vector features and the multiple multimedia resource vector features.

[0156] In this way, by training with undirected graph structures and graph neural networks, electronic devices can improve the accuracy of determining the vector features of search terms and the vector features of multimedia resources. This solves the technical problem of poor accuracy in multimedia resource searches caused by existing technologies, improves the accuracy of multimedia resource searches, and thus enriches the user's search experience.

[0157] In one embodiment, the resource information of a multimedia resource includes: number of clicks, number of search impressions, number of times the playback duration exceeds a first threshold, number of plays, and playback duration. Combined with... Figure 3 ,like Figure 4 As shown, in S302 above, the method by which the electronic device determines multiple candidate indices corresponding one-to-one with multiple multimedia resources based on the resource information of each multimedia resource specifically includes:

[0158] S401. For each multimedia resource's resource information, the electronic device performs a target operation to obtain multiple candidate indices that correspond one-to-one with the multiple multimedia resources.

[0159] The target operations are as follows: the ratio between the number of clicks and the number of search impressions is determined as the click pass rate; the ratio between the number of times the playback duration exceeds the first threshold and the number of playbacks is determined as the playback rate; and based on the click pass rate, playback rate, and playback duration, candidate indices corresponding to a multimedia resource are determined.

[0160] Among them, the number of clicks, the number of search impressions, the number of times the playback duration exceeds the first threshold, the number of plays, the playback duration, and the candidate index corresponding to a multimedia resource satisfy the following formula:

[0161] Final score=(click / show)*(long play / play)*log(play time).

[0162] Among them, Final score is the candidate index corresponding to a multimedia resource, click is the number of clicks, show is the number of search display times, long play is the number of times the playback duration is greater than the first threshold, play is the number of times it is played, and play time is the playback duration.

[0163] It should be noted that the various thresholds involved in this disclosure (such as the first threshold, the exponential threshold, etc.) can be the same value or different values, and this disclosure does not limit them.

[0164] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S401, the electronic device determines multiple candidate indices corresponding one-to-one with multiple multimedia resources based on the resource information of each multimedia resource. Target operations can be performed on the resource information of each multimedia resource to obtain multiple candidate indices corresponding one-to-one with multiple multimedia resources. A specific implementation method for determining multiple candidate indices corresponding one-to-one with multiple multimedia resources is provided, enabling the electronic device to select multiple target indices greater than the index threshold from the multiple candidate indices, determine multiple candidate multimedia resources corresponding one-to-one with the multiple target indices, and establish an undirected graph structure. This undirected graph structure improves the accuracy of determining search term vector features and multimedia resource vector features, solving the technical problem of poor accuracy in multimedia resource searches caused by existing technologies, improving the accuracy of multimedia resource searches, and thus enriching the user's search experience.

[0165] In one embodiment, combined with Figure 3 ,like Figure 5 As shown, in S305 above, the method by which the electronic device trains a graph neural network on nodes in an undirected graph structure to obtain search word vector features and multiple multimedia resource vector features specifically includes:

[0166] S501, The electronic device obtains the first adjacent node connected to the search term node via an undirected edge.

[0167] Specifically, when electronic devices train graph neural networks on nodes in an undirected graph structure to obtain search term vector features and multiple multimedia resource vector features, they can obtain the first neighboring node connected to the search term node through an undirected edge.

[0168] The first adjacent node is a multimedia resource node connected to the search term node via an undirected edge.

[0169] S502, The electronic device acquires a second adjacent node that is connected to each of the multiple multimedia resource nodes via an undirected edge.

[0170] Specifically, when electronic devices train graph neural networks on nodes in an undirected graph structure to obtain search word vector features and multiple multimedia resource vector features, they can also obtain second neighboring nodes that are connected to each of the multiple multimedia resource nodes through undirected edges.

[0171] The second adjacent node is the search term node that is connected to each multimedia resource node via an undirected edge.

[0172] S503 The electronic device performs graph neural network training on the nodes in the undirected graph structure based on the search term node, the first adjacent node, each multimedia resource node and the second adjacent node, until the undirected graph structure is trained to a convergent state, so as to obtain the search term vector features and multiple multimedia resource vector features.

[0173] Specifically, after obtaining the first neighboring node connected to the search term node through an undirected edge, and the second neighboring node connected to each of the multiple multimedia resource nodes through an undirected edge, the electronic device performs graph neural network training on the nodes in the undirected graph structure based on the search term node, the first neighboring node, each multimedia resource node, and the second neighboring node, until the undirected graph structure is trained to a convergent state, so as to obtain the search term vector features and the multiple multimedia resource vector features.

[0174] Among them, the search term node, the first adjacent node, each multimedia resource node and the second adjacent node, the search term vector feature, and the multiple multimedia resource vector features satisfy the following formula:

[0175]

[0176]

[0177] Among them, e u To search for word vector features, e i Let N be the feature of the i-th multimedia resource vector. u As the first adjacent node, N i Let be the second adjacent node, and k be the number of the kth iteration.

[0178] The above formula derivation shows that the search word vector feature e u and the feature e of the i-th multimedia resource vector i Satisfy the following formula:

[0179]

[0180]

[0181] Where K is the preset number of iterations.

[0182] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S501-S503, when the electronic device trains a graph neural network on the nodes in the undirected graph structure to obtain search term vector features and multiple multimedia resource vector features, it can obtain the first adjacent node connected to the search term node through an undirected edge, and the second adjacent node connected to each of the multiple multimedia resource nodes through an undirected edge. Subsequently, the electronic device performs graph neural network training on the nodes in the undirected graph structure based on the search term node, the first adjacent node, each multimedia resource node, and the second adjacent node, until the undirected graph structure is trained to a convergent state, so as to obtain the search term vector features and multiple multimedia resource vector features. A specific implementation method for obtaining search term vector features and multimedia resource vector features through graph neural network training is given. The electronic device can improve the accuracy of determining search term vector features and multimedia resource vector features through graph neural network training, solve the technical problem of poor accuracy in multimedia resource search caused by the prior art, improve the accuracy of multimedia resource search, and thus enrich the user's search experience.

[0183] This disclosure also provides a method for recalling multimedia resources, such as Figure 6 As shown, the multimedia resource retrieval method includes:

[0184] S601. The electronic device receives a search request that includes the target search term.

[0185] Specifically, when a user wants to browse multimedia resources of interest, they can perform a search operation on the terminal to search for those resources. In this case, the terminal responds to the user's search operation by sending a search request containing the target search terms to the electronic device. Correspondingly, the electronic device receives the search request containing the target search terms.

[0186] S602. In response to a search request, the electronic device determines at least one publishing account corresponding to the target search term from a pre-established correspondence between search terms and publishing accounts.

[0187] Specifically, upon receiving a search request that includes the target search term, the electronic device responds to the search request by determining at least one publishing account corresponding to the target search term from a pre-established mapping between search terms and publishing accounts.

[0188] In establishing the correspondence, the electronic device can obtain target vector features similar to the search term vector features of the search term from at least one multimedia category vector feature corresponding to the publishing account, and establish a correspondence between the search term and the target publishing account corresponding to the target vector feature.

[0189] S603. The electronic device sends multimedia resources published by at least one publishing account.

[0190] Specifically, after determining at least one publishing account corresponding to the target search term from the pre-established correspondence between search terms and publishing accounts, the electronic device sends the multimedia resources published by at least one publishing account.

[0191] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S601-S603, after receiving a search request including a target search term, the electronic device responds to the search request by determining at least one publishing account corresponding to the target search term from the pre-established correspondence between search terms and publishing accounts, and sending the multimedia resources published by at least one publishing account. Since the correspondence between the search term and the target publishing account corresponding to the target vector feature is established by the electronic device after obtaining target vector features similar to the search term vector features from at least one multimedia category vector feature corresponding to the publishing account, this disclosure can accurately recall multimedia resources published by the target publishing account based on the target search term, solving the technical problem of poor accuracy in multimedia resource search in the prior art, improving the accuracy of multimedia resource search, and thus enriching the user's search experience.

[0192] It is understood that, in actual implementation, the terminal / server described in the embodiments of this disclosure may include one or more hardware structures and / or software modules for implementing the aforementioned corresponding multimedia resource retrieval method. These hardware structures and / or software modules can constitute an electronic device. Those skilled in the art should readily recognize that, based on the algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, this disclosure can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0193] Based on this understanding, the present disclosure also provides a multimedia resource recall device that can be applied to electronic devices. Figure 7A A schematic diagram of the structure of a multimedia resource retrieval device provided in an embodiment of this disclosure is shown. Figure 7A As shown, the multimedia resource retrieval device may include: an acquisition unit 701 and a processing unit 702;

[0194] Acquisition unit 701 is used to acquire search terms and multiple multimedia resources; the multiple multimedia resources include live streaming multimedia resources.

[0195] The processing unit 702 is used to determine the search term vector features corresponding to the search term, and the multiple multimedia resource vector features corresponding one-to-one with multiple multimedia resources;

[0196] The processing unit 702 is also used to divide multiple multimedia resource vector features based on the publishing account corresponding to the multimedia resource to obtain a set of multimedia vector features corresponding to each publishing account.

[0197] The processing unit 702 is also used to cluster the multimedia vector feature set to obtain at least one multimedia category vector feature corresponding to the publishing account; a multimedia category vector feature is used to represent the category of multimedia resources published by an account;

[0198] The processing unit 702 is further configured to determine target vector features similar to search term vector features from at least one multimedia category vector feature, and establish a correspondence between the search term and the target publishing account corresponding to the target vector feature; the correspondence is used to recall multimedia resources published by the target publishing account based on the search term.

[0199] Optionally, the processing unit 702 is specifically used for:

[0200] Obtain the set of publishing accounts corresponding to the search terms. The set of publishing accounts includes the target publishing accounts corresponding to each target vector feature.

[0201] Using search terms as index identifiers, establish a correspondence between search terms and sets of publishing accounts.

[0202] Optionally, the processing unit 702 is specifically used for:

[0203] Obtain the distance between the search term vector features and the vector features of each multimedia category to obtain multiple distances;

[0204] Select distances that meet preset conditions from multiple distances, and determine the multimedia category vector features corresponding to distances that meet preset conditions as target vector features similar to search word vector features; preset conditions include: distance less than a distance threshold or distance ranking before at least one preset ranking.

[0205] Optionally, the processing unit 702 is specifically used for:

[0206] Retrieve resource information for each of multiple multimedia resources; the resource information is used to represent the search business information for each multimedia resource based on search terms;

[0207] Based on the resource information of each multimedia resource, multiple candidate indices are determined that correspond one-to-one with the multiple multimedia resources; the candidate indices are used to construct an undirected graph structure; the undirected graph structure includes: search term nodes corresponding to search terms, and multiple multimedia resource nodes that correspond one-to-one with the multiple multimedia resources.

[0208] From a pool of candidate indices, select multiple target indices that are greater than the index threshold, and determine multiple candidate multimedia resources that correspond one-to-one with the multiple target indices.

[0209] The search term node is connected to the multiple multimedia resource nodes corresponding to the multiple candidate multimedia resources through undirected edges to obtain an undirected graph structure;

[0210] Graph neural networks are trained on nodes in an undirected graph structure to obtain search word vector features and multiple multimedia resource vector features.

[0211] Optionally, the resource information of a multimedia resource includes: number of clicks, number of search impressions, number of times the playback duration exceeds a first threshold, number of plays, and playback duration; the processing unit 702 is specifically used for:

[0212] For each multimedia resource's resource information, the following operations are performed to obtain multiple candidate indices corresponding one-to-one with the multiple multimedia resources:

[0213] The ratio between the number of clicks and the number of search impressions is determined as the click-through rate; the ratio between the number of times the playback duration exceeds the first threshold and the number of playbacks is determined as the playback rate; based on the click-through rate, playback rate, and playback duration, candidate indices corresponding to a multimedia resource are determined.

[0214] Optionally, the processing unit 702 is specifically used for:

[0215] Find the first neighboring node that is connected to the search term node via an undirected edge;

[0216] Find the second neighboring node that is connected to each of the multiple multimedia resource nodes via an undirected edge;

[0217] Based on the search term node, the first neighboring node, each multimedia resource node and the second neighboring node, a graph neural network is trained on the nodes in the undirected graph structure until the undirected graph structure is trained to a convergent state, so as to obtain the search term vector features and multiple multimedia resource vector features.

[0218] This disclosure also provides a multimedia resource recall device that can be applied to electronic devices. Figure 7B A schematic diagram of the structure of a multimedia resource retrieval device provided in an embodiment of this disclosure is shown. Figure 7BAs shown, the multimedia resource retrieval device may include: a receiving unit 801, a processing unit 802, and a sending unit 803;

[0219] The receiving unit 801 is used to receive a search request including the target search term;

[0220] The processing unit 802 is configured to respond to a search request by determining at least one publishing account corresponding to a target search term from a pre-established correspondence between search terms and publishing accounts. When establishing the correspondence, it obtains target vector features similar to the search term vector features from at least one multimedia category vector feature corresponding to the publishing account, and establishes a correspondence between the search term and the target publishing account corresponding to the target vector feature.

[0221] Sending unit 803 is used to send multimedia resources published by at least one publishing account.

[0222] As described above, the embodiments of this disclosure can divide the electronic device into functional modules according to the above method examples. The integrated modules can be implemented in hardware or as software functional modules. Furthermore, it should be noted that the module division in these embodiments is illustrative and represents only one logical functional division; in actual implementation, other division methods may be used. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into a single processing module.

[0223] The specific methods by which each module performs its operation and the beneficial effects of the multimedia resource retrieval device in the above embodiments have been described in detail in the foregoing method embodiments, and will not be repeated here.

[0224] This disclosure also provides a terminal, which may be a user terminal such as a mobile phone or a computer. Figure 8 A schematic diagram of the structure of a terminal provided in an embodiment of this disclosure is shown. This terminal, which may be a multimedia resource retrieval device, may include at least one processor 61, a communication bus 62, a memory 63, and at least one communication interface 64.

[0225] Processor 61 can be a central processing unit (CPU), a microprocessor unit, an ASIC, or one or more integrated circuits for controlling the execution of programs according to the present disclosure. As an example, referring to Figure 7, the processing unit 702 in the electronic device performs the same functions as... Figure 8 The processor 61 in it performs the same function.

[0226] The communication bus 62 may include a path for transmitting information between the aforementioned components.

[0227] Communication interface 64 uses any transceiver-like device for communicating with other devices or communication networks, such as servers, Ethernet, radio access networks (RAN), wireless local area networks (WLAN), etc. As an example,

[0228] Memory 63 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory may exist independently and be connected to the processing unit via a bus. Memory may also be integrated with the processing unit.

[0229] The memory 63 stores the application code that executes the present invention, and its execution is controlled by the processor 61. The processor 61 executes the application code stored in the memory 63 to implement the functions of the method of the present invention.

[0230] In a specific implementation, as one example, processor 61 may include one or more CPUs, for example... Figure 8 CPU0 and CPU1 in the CPU.

[0231] In a specific implementation, as one example, the terminal may include multiple processors, for example... Figure 8 Processors 61 and 65 are included. Each of these processors may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. A processor here may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0232] In a specific implementation, as one embodiment, the terminal may further include an input device 66 and an output device 67. The input device 66 and output device 67 communicate and can accept user input in various ways. For example, the input device 66 may be a mouse, keyboard, touchscreen device, or sensing device. The output device 67 communicates with the processor 61 and can display information in various ways. For example, the output device 61 may be a liquid crystal display (LCD), a light emitting diode (LED) display device, etc.

[0233] Those skilled in the art will understand that Figure 8 The structure shown does not constitute a limitation on the terminal and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0234] This disclosure also provides a server. Figure 9 A schematic diagram of the server structure provided in this embodiment is shown. This server can be a multimedia resource retrieval device. The server can vary significantly due to differences in configuration or performance, and may include one or more processors 71 and one or more memories 72. The memory 72 stores at least one instruction, which is loaded and executed by the processor 71 to implement the multimedia resource retrieval method provided in the various method embodiments described above. Of course, the server may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server may also include other components for implementing device functions, which will not be elaborated here.

[0235] This disclosure also provides a computer-readable storage medium including instructions stored thereon, which, when executed by a processor of a computer device, enables the computer to perform the multimedia resource retrieval method provided in the embodiments described above. For example, the computer-readable storage medium may be a memory 63 including instructions, which may be executed by a processor 61 of a terminal to complete the method. As another example, the computer-readable storage medium may be a memory 72 including instructions, which may be executed by a processor 71 of a server to complete the method. Optionally, the computer-readable storage medium may be a non-transitory computer-readable storage medium, such as ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device.

[0236] This disclosure also provides a computer program product including computer instructions that, when executed on an electronic device, cause the electronic device to perform the above-described actions. Figures 1-6 The multimedia resource retrieval method shown in any of the attached figures.

[0237] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0238] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for recalling multimedia resources, characterized in that, include: Obtain the search term, multiple multimedia resources, and resource information for each of the multiple multimedia resources; The multiple multimedia resources include live streaming multimedia resources. The resource information is used to represent the search business information of each multimedia resource based on the search term; Based on the resource information of each multimedia resource, a plurality of candidate indices corresponding one-to-one with the plurality of multimedia resources are determined; The candidate indices are used to construct the undirected graph structure; The undirected graph structure includes: a search term node corresponding to the search term, and multiple multimedia resource nodes corresponding one-to-one with the multiple multimedia resources; From the plurality of candidate indices, select a plurality of target indices that are greater than the index threshold, and determine a plurality of candidate multimedia resources that correspond one-to-one with the plurality of target indices; The search term node is connected to the multiple multimedia resource nodes corresponding to the multiple candidate multimedia resources through undirected edges to obtain the undirected graph structure; Graph neural network is trained on the nodes in the undirected graph structure to obtain the search term vector features corresponding to the search term, and the multiple multimedia resource vector features corresponding one-to-one with the multiple multimedia resources. Based on the publishing account corresponding to the multimedia resource, the vector features of the multiple multimedia resources are divided to obtain the set of multimedia vector features corresponding to each publishing account; Clustering is performed on the multimedia vector feature set to obtain at least one multimedia category vector feature corresponding to the publishing account; a multimedia category vector feature is used to represent the category of multimedia resources published by an account; From the at least one multimedia category vector feature, a target vector feature similar to the search term vector feature is determined, and a correspondence is established between the search term and the target publishing account corresponding to the target vector feature; the correspondence is used to recall multimedia resources published by the target publishing account based on the search term.

2. The multimedia resource retrieval method according to claim 1, characterized in that, The step of establishing the correspondence between the search term and the target publishing account corresponding to the target vector feature includes: Obtain the set of publishing accounts corresponding to the search term, wherein the set of publishing accounts includes the target publishing accounts corresponding to each of the target vector features; Using the search terms as index identifiers, establish a correspondence between the search terms and the set of publishing accounts.

3. The multimedia resource retrieval method according to claim 1, characterized in that, Determining target vector features similar to the search term vector features from the at least one multimedia category vector features includes: Obtain the distance between the search term vector features and each multimedia category vector feature to obtain multiple distances; Distances that satisfy preset conditions are selected from the plurality of distances, and the multimedia category vector features corresponding to the distances that satisfy the preset conditions are determined as target vector features similar to the search term vector features; the preset conditions include: the distance is less than a distance threshold or the distance is ranked before at least one of the preset rankings.

4. The multimedia resource retrieval method according to claim 1, characterized in that, The resource information of a multimedia resource includes: number of clicks, number of search impressions, number of times the playback duration exceeds a first threshold, number of plays, and playback duration; based on the resource information of each multimedia resource, multiple candidate indices corresponding one-to-one with the multiple multimedia resources are determined, including: For each multimedia resource's resource information, the following operations are performed to obtain multiple candidate indices corresponding one-to-one with the multiple multimedia resources: The ratio between the number of clicks and the number of search impressions is determined as the click pass rate; the ratio between the number of times the playback duration exceeds the first threshold and the number of playbacks is determined as the playback rate; based on the click pass rate, the playback rate, and the playback duration, a candidate index corresponding to the multimedia resource is determined.

5. The multimedia resource retrieval method according to claim 4, characterized in that, The step of training a graph neural network on the nodes in the undirected graph structure to obtain the search word vector features and the multiple multimedia resource vector features includes: Obtain the first neighboring node connected to the search term node via the undirected edge; Obtain the second neighboring node that is connected to each of the plurality of multimedia resource nodes through the undirected edge; Based on the search term node, the first neighboring node, each multimedia resource node, and the second neighboring node, the graph neural network is trained on the nodes in the undirected graph structure until the undirected graph structure is trained to a convergent state, so as to obtain the search term vector features and the multiple multimedia resource vector features.

6. A method for recalling multimedia resources, characterized in that, include: Receive search requests that include the target search term; In response to the search request, at least one publishing account corresponding to the target search term is determined from a pre-established correspondence between search terms and publishing accounts. When establishing the correspondence, target vector features similar to the search term vector features are obtained from at least one multimedia category vector feature corresponding to the publishing account, and a correspondence is established between the search term and the target publishing account corresponding to the target vector feature. The search term vector features and the multimedia resource vector features are obtained by training a graph neural network on nodes in an undirected graph structure. The undirected graph structure includes: a search term node corresponding to the search term, and multiple multimedia resource nodes corresponding one-to-one with multiple multimedia resources. The undirected graph structure is obtained by: determining multiple candidate indices corresponding one-to-one with each of the multiple multimedia resources based on the resource information of each multimedia resource; selecting multiple target indices greater than an index threshold from the multiple candidate indices, and determining multiple candidate multimedia resources corresponding one-to-one with the multiple target indices; connecting the search term node and the multiple multimedia resource nodes corresponding to the multiple candidate multimedia resources through undirected edges to obtain the undirected graph structure. Send multimedia resources published by at least one publishing account.

7. A multimedia resource retrieval device, characterized in that, include: Acquisition unit and processing unit; The acquisition unit is used to acquire search terms, multiple multimedia resources, and resource information of each multimedia resource among the multiple multimedia resources; The multiple multimedia resources include live streaming multimedia resources. The resource information is used to represent the search business information of each multimedia resource based on the search term; The processing unit is used to determine multiple candidate indices that correspond one-to-one with the multiple multimedia resources based on the resource information of each multimedia resource. The candidate indices are used to construct the undirected graph structure; The undirected graph structure includes: a search term node corresponding to the search term, and multiple multimedia resource nodes corresponding one-to-one with the multiple multimedia resources; The processing unit is further configured to select multiple target indices that are greater than the index threshold from the multiple candidate indices, and determine multiple candidate multimedia resources that correspond one-to-one with the multiple target indices; The processing unit is further configured to connect the search term node with the multiple multimedia resource nodes corresponding to the multiple candidate multimedia resources through undirected edges to obtain the undirected graph structure. The processing unit is further configured to train a graph neural network on the nodes in the undirected graph structure to obtain search term vector features corresponding to the search term, and multiple multimedia resource vector features corresponding one-to-one with the multiple multimedia resources. The processing unit is further configured to divide the vector features of the multiple multimedia resources based on the publishing account corresponding to the multimedia resources, and obtain a set of multimedia vector features corresponding to each publishing account. The processing unit is further configured to cluster the multimedia vector feature set to obtain at least one multimedia category vector feature corresponding to the publishing account; a multimedia category vector feature is used to represent the category of multimedia resources published by an account. The processing unit is further configured to determine, from the at least one multimedia category vector feature, a target vector feature similar to the search term vector feature, and establish a correspondence between the search term and the target publishing account corresponding to the target vector feature; the correspondence is used to recall multimedia resources published by the target publishing account based on the search term.

8. The multimedia resource retrieval device according to claim 7, characterized in that, The processing unit is specifically used for: Obtain the set of publishing accounts corresponding to the search term, wherein the set of publishing accounts includes the target publishing accounts corresponding to each of the target vector features; Using the search terms as index identifiers, establish a correspondence between the search terms and the set of publishing accounts.

9. The multimedia resource retrieval device according to claim 7, characterized in that, The processing unit is specifically used for: Obtain the distance between the search term vector features and each multimedia category vector feature to obtain multiple distances; Select a distance that meets a preset condition from the plurality of distances, and determine the multimedia category vector feature corresponding to the distance that meets the preset condition as a target vector feature similar to the search term vector feature; The preset conditions include: the distance is less than a distance threshold or the distance is sorted before at least one preset sorting condition.

10. The multimedia resource retrieval device according to claim 7, characterized in that, Resource information for a multimedia resource includes: number of clicks, number of search impressions, number of times playback duration exceeds a first threshold, number of plays, and playback duration; the processing unit is specifically used for: For each multimedia resource's resource information, the following operations are performed to obtain multiple candidate indices corresponding one-to-one with the multiple multimedia resources: The ratio between the number of clicks and the number of search impressions is determined as the click pass rate; the ratio between the number of times the playback duration exceeds the first threshold and the number of playbacks is determined as the playback rate; based on the click pass rate, the playback rate, and the playback duration, a candidate index corresponding to the multimedia resource is determined.

11. The multimedia resource retrieval device according to claim 10, characterized in that, The processing unit is specifically used for: Obtain the first neighboring node connected to the search term node via the undirected edge; Obtain the second neighboring node that is connected to each of the plurality of multimedia resource nodes through the undirected edge; Based on the search term node, the first neighboring node, each multimedia resource node, and the second neighboring node, the graph neural network is trained on the nodes in the undirected graph structure until the undirected graph structure is trained to a convergent state, so as to obtain the search term vector features and the multiple multimedia resource vector features.

12. A multimedia resource retrieval device, characterized in that, include: Receiving unit, processing unit, and transmitting unit; The receiving unit is used to receive a search request including the target search term; The processing unit is configured to, in response to the search request, determine at least one publishing account corresponding to the target search term from a pre-established correspondence between search terms and publishing accounts. Specifically, when establishing the correspondence, target vector features similar to the search term vector features of the search term are obtained from at least one multimedia category vector feature corresponding to the publishing account, and a correspondence is established between the search term and the target publishing account corresponding to the target vector feature. The search term vector features and the multimedia resource vector features are obtained by training a graph neural network on nodes in an undirected graph structure. The undirected graph structure includes: a search term node corresponding to the search term, and multiple multimedia resource nodes corresponding one-to-one with multiple multimedia resources. The undirected graph structure is obtained by: determining multiple candidate indices corresponding one-to-one with each of the multiple multimedia resources based on the resource information of each multimedia resource; selecting multiple target indices greater than an index threshold from the multiple candidate indices, and determining multiple candidate multimedia resources corresponding one-to-one with the multiple target indices; connecting the search term node and the multiple multimedia resource nodes corresponding to the multiple candidate multimedia resources through undirected edges to obtain the undirected graph structure. The sending unit is used to send multimedia resources published by the at least one publishing account.

13. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the multimedia resource retrieval method as described in any one of claims 1-5 or 6.

14. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the multimedia resource retrieval method as described in any one of claims 1-5 or 6.

15. A computer program product, comprising instructions, characterized in that, When the instruction is executed on an electronic device, the electronic device performs the multimedia resource recall method as described in any one of claims 1-5 or 6.

Images