Query information rewriting model training method, music searching method, equipment and medium

Abstract

The invention relates to a query information rewriting model training method, a music searching method, equipment and a medium. The training method comprises the following steps: acquiring sample music query information of a sample user and user preference of the sample user for a sample music search result; the sample music query information carries emoticon information, and the sample music search result is a search result obtained by searching through the sample music query information; inputting the sample music query information and the user preference into a query information rewriting model to be trained, and rewriting emoticon information carried in the sample music query information through the query information rewriting model to obtain prediction query rewriting information corresponding to the sample music query information; and training the query information rewriting model by using the difference between the predicted query rewriting information and the actual query rewriting information to obtain a trained query information rewriting model. By adopting the method, the application range of the query information rewriting model can be expanded.

Description

Technical Field

[0001] This application relates to the field of search technology, and in particular to a method for training a query information rewriting model, a music search method, computer equipment, storage media, and computer program products. Background Art

[0002] With the development of search technology, a technology has emerged that uses search models to achieve multimedia search, such as music search. Users can enter query information in the search box to perform a music search, and the search model can then complete the music search based on the query information.

[0003] Since the query information entered by users may not be accurate enough, a rewriting model is usually used to rewrite the query information entered by users to improve the user's search intent. The rewritten query information is then input into the search model to realize the music search. This method can effectively improve the accuracy of music search.

[0004] However, current query information rewriting techniques are generally only applicable to rewriting text entered by users, thus limiting the applicability of query information rewriting models. Summary of the Invention

[0005] Therefore, it is necessary to provide a query information rewriting model training method, music search method, device, computer equipment, computer-readable storage medium, and computer program product that can improve the applicability of the query information rewriting model in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a method for training a query information rewriting model, including:

[0007] Obtain sample music query information from sample users, as well as the user preferences of the sample users for sample music search results; wherein, the sample music query information carries emoji information, and the sample music search results are music search results obtained through the sample music query information;

[0008] The sample music query information and the user preference are input into the query information rewriting model to be trained. The emoji information carried in the sample music query information is rewritten by the query information rewriting model to obtain the predicted query rewriting information corresponding to the sample music query information.

[0009] By utilizing the difference between the predicted query rewriting information and the actual query rewriting information corresponding to the sample music query information, the query information rewriting model is trained to obtain the trained query information rewriting model.

[0010] In one embodiment, the step of rewriting the emoji information carried in the sample music query information using the query information rewriting model to obtain the predicted query rewriting information corresponding to the sample music query information includes: mapping the emoji information using the query information rewriting model to obtain initial query rewriting information; merging the initial query rewriting information with the sample music query information to obtain candidate query rewriting information; and obtaining the predicted query rewriting information from the candidate query rewriting information according to the user preference.

[0011] In one embodiment, obtaining the predicted query rewriting information from the candidate query rewriting information according to the user preference includes: sorting the candidate query rewriting information according to the user preference to obtain a sorting result of the candidate query rewriting information; and obtaining a preset number of candidate query rewriting information based on the sorting result as the predicted query rewriting information.

[0012] In one embodiment, obtaining the sample music query information of the sample user and the user preferences of the sample user for the sample music search results includes: obtaining a pre-built search behavior log; obtaining from the search behavior log the sample user's historical music query information used for searching sample music, the sample user's user interaction results for the historical music search results, and the sample user's user preference settings; wherein the historical music search results are music search results obtained through the historical music query information; using the historical music query information as the sample music query information, and using the user interaction results and the user preference settings as the user preferences.

[0013] In one embodiment, before obtaining the pre-built search behavior log, the method further includes: in response to the sample user's music search request for sample music, obtaining the sample user's historical music query information used to search for sample music, and the historical music search results corresponding to the historical music query information; obtaining the sample user's user interaction results for the historical music search results, and the sample user's user preference settings; and constructing the search behavior log using the historical music query information, the historical music search results, the user interaction results, and the user preference settings.

[0014] Secondly, this application also provides a music search method, including:

[0015] In response to a target user's music search request for a target music, the system obtains the music query information input by the target user for searching the target music, as well as the target user's first user preference corresponding to historical music search requests; the music query information carries emoji information.

[0016] The music query information and the query information rewriting model trained by the first user preference input are used to rewrite the emoji information carried in the music query information to obtain query rewritten information corresponding to the music query information; wherein, the query information rewriting model is trained by the query information rewriting model training method as described in any embodiment of the first aspect;

[0017] The rewritten query information is input into the trained music search model, and music search results for the target music are obtained through the music search model.

[0018] In one embodiment, after obtaining the music search results corresponding to the music query information through the music search model, the method further includes: obtaining the target user's second user preference for the music search results; constructing a search behavior log of the target user for the target music based on the music query information and the second user preference; the search behavior log is used to obtain a new first user preference when a new music search request is received from the target user again; wherein the new first user preference includes the second user preference.

[0019] Thirdly, this application also provides a query information rewriting model training device, comprising:

[0020] The sample information acquisition module is used to acquire sample music query information of sample users, as well as the user preferences of the sample users for sample music search results; wherein, the sample music query information carries emoticon information, and the sample music search results are music search results obtained through the sample music query information;

[0021] The prediction rewriting acquisition module is used to input the sample music query information and the user preference into the query information rewriting model to be trained, and rewrite the emoji information carried in the sample music query information through the query information rewriting model to obtain the prediction query rewriting information corresponding to the sample music query information.

[0022] The rewriting model training module is used to train the query information rewriting model by utilizing the difference between the predicted query rewriting information and the actual query rewriting information corresponding to the sample music query information, so as to obtain the trained query information rewriting model.

[0023] Fourthly, this application also provides a music search device, comprising:

[0024] The search request response module is used to respond to a target user's music search request for a target music, obtain the music query information input by the target user for searching the target music, and the target user's first user preference corresponding to historical music search requests; the music query information carries emoji information;

[0025] The query information rewriting module is used to rewrite the emoji information carried in the music query information and the query information rewriting model trained by the first user preference input, thereby obtaining query rewritten information corresponding to the music query information; wherein, the query information rewriting model is trained by the query information rewriting model training method as described in any embodiment of the first aspect;

[0026] The search result acquisition module is used to input the rewritten query information into the trained music search model, and obtain music search results for the target music through the music search model.

[0027] Fifthly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any one of the embodiments of the first or second aspect.

[0028] In a sixth aspect, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any one of the embodiments of the first or second aspect.

[0029] In a seventh aspect, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in any one of the embodiments of the first or second aspect.

[0030] The aforementioned query information rewriting model training method, music search method, device, computer equipment, storage medium, and computer program product acquire sample music query information from sample users and their user preferences for sample music search results. The sample music query information carries emoji information, and the sample music search results are music search results obtained using the sample music query information. The sample music query information and user preferences are input into the query information rewriting model to be trained. The model rewrites the emoji information carried in the sample music query information to obtain the predicted query rewriting information corresponding to the sample music query information. The difference between the predicted query rewriting information and the actual query rewriting information corresponding to the sample music query information is used to train the query information rewriting model, resulting in a trained query information rewriting model. This application collects sample music query information containing emoticons input by sample users when initiating music searches, as well as user preferences for sample music search results obtained through the sample music query information. This sample music query information and user preferences are then input into a query information rewriting model to be trained. This model can rewrite the emoticons carried in the sample music query information to obtain corresponding predicted query rewriting information. The difference between the predicted and actual query rewriting information can then be used to train the query information rewriting model. The query information rewriting model trained in this way can rewrite music query information carrying emoticon information, and is not limited to text rewriting, thus improving the applicability of the query information rewriting model. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart illustrating the method for rewriting the model training process based on query information in one embodiment.

[0033] Figure 2 This is a schematic diagram of the process for obtaining predicted query rewrite information in one embodiment;

[0034] Figure 3 This is a flowchart illustrating the process of obtaining sample music query information and user preferences in one embodiment;

[0035] Figure 4 This is a flowchart illustrating a music search method in one embodiment;

[0036] Figure 5 This is a schematic diagram of an emoji search method based on active transformation search logs in one embodiment;

[0037] Figure 6 This is a flowchart illustrating the process of building a log collection module in one embodiment;

[0038] Figure 7 A flowchart illustrating the search model rewriting by actively changing the emojis in the search log in one embodiment;

[0039] Figure 8 This is an interactive schematic diagram of a music search method in one embodiment;

[0040] Figure 9 Here is a structural block diagram of a query information rewriting model training device in one embodiment;

[0041] Figure 10 This is a structural block diagram of a music search device in one embodiment;

[0042] Figure 11 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0044] In one embodiment, such as Figure 1 As shown, a method for training a query information rewriting model is provided. This embodiment illustrates the method by applying it to a server. It is understood that this method can also be applied to a terminal, or to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0045] Step S101: Obtain sample music query information of sample users and user preferences of sample users for sample music search results; wherein, the sample music query information carries emoticon information, and the sample music search results are music search results obtained through the sample music query information.

[0046] Among them, the sample music query information refers to the music query information used to train the query information rewriting model. This query information can be the music query information entered by the sample user when performing historical music search behavior. This query information can contain emoticons, such as emoji emoticons. The music search results refer to the search results obtained after searching for music by entering the sample music query information. User preferences are the sample user's preferences for the sample music search results. These preferences can be represented by the sample user's interactive behavior towards the sample music search results, such as browsing time, clicks, likes, favorites, and playback.

[0047] Specifically, when training a query information rewriting model, you can first collect sample music query information that carries emojis, as well as the user preferences of sample users for the sample music search results obtained from the sample music query information, as training samples for the model to train a query information rewriting model that rewrites emojis.

[0048] Step S102: The query information rewriting model is trained by rewriting the sample music query information and the user preference input. The emoji information carried in the sample music query information is rewritten by the query information rewriting model to obtain the predicted query rewriting information corresponding to the sample music query information.

[0049] Predictive query rewriting information is the predicted rewritten word obtained by rewriting the emoji information carried in the sample music query information by the query information rewriting model. Specifically, after obtaining the sample music query information and user preferences, the sample music query information and user preferences can be input into the query information rewriting model to be trained. The query information rewriting model will then rewrite the emoji information carried in the sample music query information to output the corresponding predicted query rewriting information.

[0050] Step S103: Using the difference between the predicted query rewriting information and the actual query rewriting information corresponding to the sample music query information, train the query information rewriting model to obtain the trained query information rewriting model.

[0051] The actual query rewrite information is the real query rewrite information corresponding to the sample music query information. This query rewrite information can be obtained by users annotating the sample music query information, such as the "..." in the sample music query information. "Can be rewritten as "music" or "song", while " "In a birthday context, it can be rewritten as "Happy Birthday," and in a celebration context, it can be rewritten as "Celebrate," and so on.

[0052] Specifically, after obtaining the predicted query rewriting information corresponding to the sample music query information through the query information rewriting model, the server can also use the difference between the actual query rewriting information corresponding to the sample music query information and the above sample music query information to modify the query information rewriting model, thereby obtaining the trained query information rewriting model. Since the model has learned the rewriting of sample music query information carrying emoji information, it can realize the rewriting of emojis.

[0053] In the above-mentioned query information rewriting model training method, sample music query information of sample users and user preferences for sample music search results are obtained. The sample music query information carries emoji information, and the sample music search results are music search results obtained through the sample music query information. The sample music query information and user preferences are input into the query information rewriting model to be trained. The query information rewriting model rewrites the emoji information carried in the sample music query information to obtain the predicted query rewriting information corresponding to the sample music query information. The difference between the predicted query rewriting information and the actual query rewriting information corresponding to the sample music query information is used to train the query information rewriting model to obtain the trained query information rewriting model. This application collects sample music query information containing emoticons input by sample users when initiating music searches, as well as user preferences for sample music search results obtained through the sample music query information. This sample music query information and user preferences are then input into a query information rewriting model to be trained. This model can rewrite the emoticons carried in the sample music query information to obtain corresponding predicted query rewriting information. The difference between the predicted and actual query rewriting information can then be used to train the query information rewriting model. The query information rewriting model trained in this way can rewrite music query information carrying emoticon information, and is not limited to text rewriting, thus improving the applicability of the query information rewriting model.

[0054] In one embodiment, such as Figure 2 As shown, step S102 may further include:

[0055] Step S201: Map the emoji information using the query information rewriting model to obtain the initial query rewriting information.

[0056] The initial query rewriting information refers to the query rewriting information directly obtained by the query information rewriting model through mapping the emoji information. In this embodiment, after the server rewrites the sample music query information and the user preference input query information, the query information rewriting model can first extract the emoji information carried in the sample music query information, and then map the emoji information as the query term obtained by the model mapping, which is the initial query rewriting information.

[0057] Step S202: Merge the initial query rewrite information with the sample music query information to obtain candidate query rewrite information;

[0058] Step S203: Based on user preferences, obtain predicted query rewriting information from candidate query rewriting information.

[0059] Candidate query rewriting information refers to the query terms that can be used as the predicted query rewriting information output by the query information rewriting model. The query terms consist of two parts: one part is the query terms initially input by the sample user, i.e., the sample music query information, and the other part is the query terms mapped by the query information rewriting model, i.e., the initial query rewriting information.

[0060] Specifically, after the server obtains the initial query rewriting information, it can merge the initial query rewriting information with the sample music query information input by the sample user to obtain candidate query rewriting information. Then, the query rewriting model can further filter the candidate query rewriting information according to user preferences to obtain predictive query rewriting information that better meets user preferences.

[0061] In this embodiment, after rewriting the model with the sample music query information and the user preference input query information, the query information rewriting model can first map the emoticons in the sample music query information to obtain the initial query rewriting information. Then, the initial query rewriting information can be merged with the sample music query information to obtain the candidate query rewriting information. Finally, the user preference is used to filter the predicted query rewriting information from the candidate query rewriting information. This method can ensure the completeness and accuracy of the predicted query rewriting information.

[0062] Furthermore, step S203 may further include: sorting the candidate query rewriting information according to user preferences to obtain a sorting result of the candidate query rewriting information; and obtaining a preset number of candidate query rewriting information based on the sorting result as predicted query rewriting information.

[0063] In this embodiment, the predicted query rewriting information can be obtained by sorting the candidate query rewriting information according to user preferences. For example, the candidate query rewriting information can first be sorted by user preference analysis and relevance ranking, and then the candidate query rewriting information can be sorted in a personalized way to obtain the ranking result of each candidate query rewriting information. The higher the ranking, the more the candidate query rewriting information is in line with user preferences.

[0064] Then, the server can further obtain a preset number of candidate query rewriting information based on the sorting results, as predicted query rewriting information. For example, the top preset number of candidate query rewriting information in the sorting results can be used as predicted query rewriting information.

[0065] In this embodiment, user preferences can also be used to sort the candidate query rewriting information to obtain the sorting result of the candidate query rewriting information. Based on the sorting result, the predicted query rewriting information can be filtered. In this way, the filtered predicted query rewriting information can better meet user preferences and further improve the filtering accuracy of the predicted query rewriting information.

[0066] In one embodiment, such as Figure 3 As shown, step S101 may further include:

[0067] Step S301: Obtain the pre-built search behavior log.

[0068] Among them, the search behavior log is used to store information related to a user's historical music search behavior. When a user performs a music search, the information related to that behavior can be packaged into a log and stored. When training the query information rewriting model, the pre-built search behavior log can be used as a training sample to train the model.

[0069] Step S302: Obtain from the search behavior log the sample user's historical music query information used to search for sample music, the sample user's user interaction results for historical music search results, and the sample user's user preference settings; wherein the historical music search results are the music search results obtained through the historical music query information;

[0070] Step S303: Use historical music query information as sample music query information, and use user interaction results and user preference settings as user preferences.

[0071] Sample music can refer to the music that sample users intended to search for during their historical music search behavior. Historical music query information refers to the music query information entered by sample users when they conducted historical music search behavior. User interaction results of sample users on historical music search results refer to the user interaction results of sample users on music search results obtained by using historical music query information. This may include the browsing time, clicks, likes, favorites, playback and other interactive behaviors of sample users on historical music search results. User preference settings refer to the past preference settings of sample users, etc.

[0072] Specifically, the search behavior log can pre-store information related to the sample user's historical music search behavior for the sample music. This may include the sample user's historical music query information used to search for the sample music, the sample user's user interaction results with the historical music search results, and the sample user's user preference settings. Furthermore, the sample user's historical music query information used to search for the sample music can be used as the sample music query information, and the sample user's user interaction results with the historical music search results and the sample user's user preference settings can be used as the sample user's user preferences for the sample music search results.

[0073] In this embodiment, the historical music query information of the sample user used to search for sample music, the user interaction results of the sample user for the historical music search results, and the user preference settings of the sample user can also be obtained from the pre-built search behavior log. Thus, the historical music query information can be used as the sample music query information, and the user interaction results and user preference settings can be used as the user preferences. This method can improve the accuracy of obtaining the sample music query information and user preferences.

[0074] Furthermore, prior to step S301, the method may include: responding to a sample user's music search request for sample music, obtaining the sample user's historical music query information used to search for sample music, and the historical music search results corresponding to the historical music query information; obtaining the sample user's user interaction results for the historical music search results, and the sample user's user preference settings; and constructing a search behavior log using the historical music query information, historical music search results, user interaction results, and user preference settings.

[0075] A music search request refers to a sample user's historical search requests for the sample music. Historical music query information refers to the music query information entered by the sample user when initiating a music search request for the sample music, while historical music search results refer to the search results obtained by the sample user when initiating a music search request for the sample music. Specifically, when a sample user initiates a music search request for the sample music, the sample user's historical music query information and the historical music search results obtained using that information can be collected. Furthermore, the sample user's interaction results with the historical music search results and their user preference settings can also be collected. Using these historical music query information, historical music search results, user interaction results, and user preference settings, a search behavior log corresponding to that music search request can be constructed.

[0076] In this embodiment, when a sample user initiates a music search request for sample music, the sample user's historical music query information used to search for sample music, as well as the historical music search results obtained through the historical music query information, can also be collected. Furthermore, the sample user's user interaction results with the historical music search results and the sample user's preference settings can also be collected. In this way, the above information can be used to construct a search behavior log. The search behavior log constructed in this way can contain more detailed information, thus providing a solid foundation for training the query information rewriting model.

[0077] In one embodiment, such as Figure 4 As shown, a music search method is provided. This embodiment illustrates the method applied to a server. It is understood that this method can also be applied to a terminal, or to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0078] Step S401: In response to the target user's music search request for the target music, obtain the music query information input by the target user for searching the target music, and the target user's first user preference corresponding to the historical music search requests; the music query information carries emoji information.

[0079] The target user refers to the user who needs to search for music, and the target music refers to the music that the target user needs to search for. When the target user needs to search for music, they can send a music search request to the server for the target music. The server can then respond to the request and obtain the music query information used by the target user to search for the target music. This music query information may include emoticons and the user preferences corresponding to the target user's historical music search requests, i.e., the first user preference. This first user preference can be obtained through the target user's search behavior log.

[0080] Specifically, when a target user initiates a music search request for the target music, the server can also obtain the music query information containing emoticons entered by the target user to search for the target music, and the server can also query the search behavior log that stores the target user's historical music search behavior information to obtain the target user's first user preference corresponding to the historical music search request.

[0081] Step S402: The query information rewriting model trained with the music query information and the first user preference input is used to rewrite the emoji information carried in the music query information to obtain the query rewritten information corresponding to the music query information; wherein, the query information rewriting model is trained by the query information rewriting model training method as described in any of the above embodiments.

[0082] The trained query information rewriting model refers to the query information rewriting model trained using the query information rewriting model training method described in any of the preceding embodiments. The rewritten query information refers to the query keyword information generated after rewriting the emoji information carried in the music query information. Specifically, after obtaining the music query information and the first user preference, the server can input the music query information and the first user preference into the trained query information rewriting model. The query information rewriting model then rewrites the emoji information carried in the music query information to obtain the corresponding rewritten query information.

[0083] The rewriting process can begin by mapping the emojis carried in the music query information to obtain initial rewritten information. Then, the initial rewritten information is merged with the original music query information. Finally, the merged query information is sorted using the first user preference to filter out a preset number of query information as the query rewritten information corresponding to the music query information.

[0084] Step S403: Input the rewritten query information into the trained music search model, and obtain music search results for the target music through the music search model.

[0085] After obtaining the query rewrite information, it can also be input into a pre-trained music search model, which will then generate music search results for the target music.

[0086] In the aforementioned music search method, in response to a target user's music search request for a target music, the method obtains the music query information input by the target user for searching the target music, as well as the target user's first user preference corresponding to historical music search requests; the music query information carries emoji information; the music query information and the first user preference are input into a trained query information rewriting model, and the emoji information carried in the music query information is rewritten by the query information rewriting model to obtain query rewritten information corresponding to the music query information; wherein, the query information rewriting model is trained by the query information rewriting model training method described in any of the preceding embodiments; the query rewritten information is input into the trained music search model, and the music search model obtains music search results for the target music. This embodiment uses a music query information containing emoticons, input by the target user to search for target music, and a query information rewriting model trained based on the target user's first user preference input. This model rewrites the emoticons in the music query information to obtain rewritten query information. This rewritten query information is then input into a trained music search model to complete the music search for the target music. Since the query information rewriting model used in this embodiment can rewrite music query information containing emoticons, and is not limited to text rewriting, it is applicable to searching music query information containing emoticons, thus improving search accuracy.

[0087] In one embodiment, after step S403, the method may further include: obtaining the target user's second user preference for music search results; constructing a search behavior log of the target user for the target music based on the music query information and the second user preference; the search behavior log is used to obtain a new first user preference when a new music search request is received from the target user again; wherein the new first user preference includes the second user preference.

[0088] The second user preference refers to the user preference of the target user in response to the music search results obtained from the music search request. In this embodiment, after completing the music search, the server can also collect music search behavior information corresponding to the music search request initiated by the target user for the target music to construct a corresponding search behavior log. This log can record the music query information entered by the target user in the music search request for the target music, as well as the target user's second user preference for the music search results. Furthermore, this log can be used to extract the stored second user preference when the target user initiates another music search request, and use it as part of the first user preference extracted when the target user initiates a new music search request.

[0089] In this embodiment, the second user preference of the target user for music search results can also be collected to construct the target user's search behavior log for the target music. When a new music search request is received from the target user again, the stored second user preference can be extracted from the log as part of the first user preference used to rewrite the model for input query information. This method can further improve the accuracy of music search.

[0090] In one embodiment, an emoji search method based on proactively transformed search logs is also provided. This method, based on intelligent search improvement methods using user behavior analysis, identifies adjustments made by users after their initial search fails to yield satisfactory results by capturing and analyzing their secondary search behavior. The system records the user's initial emoji input for searching and associates the two when the user subsequently enters text keywords for searching again. Through conditional filtering and machine learning algorithms, emoji query rewrite pairs with high confidence are extracted, and the emojis are automatically converted into corresponding text content, thereby providing more accurate search results. The main features are as follows:

[0091] 1. The system is equipped with a dedicated log collection module that captures user search behavior in real time, recording initial query terms, secondary query terms, timestamps, and contextual information (such as geographical location and device type). These log data are stored in a high-performance distributed storage system to support rapid data extraction and analysis.

[0092] 2. Advanced machine learning algorithms and large language models are used to train on user behavior data to identify high-confidence emoji query rewriting pairs. The model can ensure the accuracy of the rewriting based on different scenarios and semantic differences. For example, " "Can be rewritten as "music" or "song", while " In the context of a birthday, it should be rewritten as "Happy Birthday," and in the context of a celebration, it should be rewritten as "Celebrate."

[0093] 3. Introduce a user feedback mechanism. When users click or listen to the rewritten search results, the system collects this feedback data to optimize future rewriting rules and models, enabling the system to continuously learn and improve.

[0094] Among them, such as Figure 5 As shown, the specific implementation process may include:

[0095] 1. When a user enters a query in the search box, the system collects the user's search content, the results of the first search, the content of the second search, the results of the second search, the user's actions on the first and second search results (viewing time, clicks, likes, favorites, plays, etc.), the duration between the two search actions, the user's past preference settings, and other behaviors, provided that the user authorizes the system to do so.

[0096] 2. After the backend recognizes that the user's input is related to emoji, it inputs the user's search behavior into the "emoji rewriting model based on active change search log" to obtain the rewritten query;

[0097] 3. Search the rewritten query and retrieve the results, then insert one or more results into the app's search results page.

[0098] The specific process for each step is as follows:

[0099] 1. Log collection module, such as Figure 6 As shown, by cleaning user-initiated search transformation logs, active transformation pairs containing emoji content with high confidence are mined and used as training samples for the model.

[0100] 2. Build a neural network-based model using a deep learning framework (such as TensorFlow or PyTorch), such as an LSTM or Transformer model. Use the extracted query rewrite pairs as training data, with emojis as input features and the corresponding actual query terms as the target output, to train the model. By continuously learning the semantic relationships in this data, the model gradually masters the mapping rules between emojis and various query intentions. This allows it to predict the query terms that best match the user's intent when faced with new emoji queries, and generate more accurate search results accordingly.

[0101] 3. Recall and Ranking Module. When the system receives another search query containing emojis from the user, the recall module will retrieve the results of both the original query and the model-mapped query. The ranking module will then rank the retrieved results based on relevance and personalization according to the user's past preferences, returning the top n results.

[0102] The process of generating rewrite pairs from the model can be described as follows: Figure 7 As shown.

[0103] After the model training is complete, it can be used to perform music search. The specific process of music search is as follows: Figure 8 As shown, when a user enters a query containing emojis, user logs can be collected first, and the model can extract and return the rewritten pairs of the query. Then, the rewritten pairs are input into the music search model to complete the music search.

[0104] This embodiment innovatively designs a process for emoji search, utilizing user search logs and an emoji rewriting model based on proactively modified search logs to expand recall records, enhance emoji compatibility, accurately target user content, and optimize recommendation results. It also enhances personalized experience and improves the satisfaction of streaming media software.

[0105] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0106] Based on the same inventive concept, this application also provides a query information rewriting model training device for implementing the query information rewriting model training method described above, and a music search device for implementing the music search method described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more query information rewriting model training devices and music search devices provided below can be found in the limitations of the query information rewriting model training method and music search method described above, and will not be repeated here.

[0107] In one embodiment, such as Figure 9 As shown, a query information rewriting model training device is provided, including: a sample information acquisition module 901, a prediction rewriting acquisition module 902, and a rewriting model training module 903, wherein:

[0108] The sample information acquisition module 901 is used to acquire sample music query information of sample users, as well as user preferences of sample users for sample music search results; wherein, the sample music query information carries emoticon information, and the sample music search results are music search results obtained through the sample music query information;

[0109] The prediction rewriting acquisition module 902 is used to input the sample music query information and user preferences into the query information rewriting model to be trained. The query information rewriting model rewrites the emoji information carried in the sample music query information to obtain the predicted query rewriting information corresponding to the sample music query information.

[0110] The rewriting model training module 903 is used to train the query information rewriting model by utilizing the difference between the predicted query rewriting information and the actual query rewriting information corresponding to the sample music query information, so as to obtain the trained query information rewriting model.

[0111] In one embodiment, such as Figure 10 As shown, a music search device is provided, including: a search request response module 1001, a query information rewriting module 1002, and a search result acquisition module 1003, wherein:

[0112] The search request response module 1001 is used to respond to a target user's music search request for a target music, obtain the music query information input by the target user for searching for the target music, and the target user's first user preference corresponding to historical music search requests; the music query information carries emoticon information;

[0113] The query information rewriting module 1002 is used to rewrite the emoji information carried in the music query information and the query information rewriting model trained by the first user preference input, so as to obtain the query rewritten information corresponding to the music query information; wherein, the query information rewriting model is trained by the query information rewriting model training method as described in any of the above embodiments.

[0114] The search result acquisition module 1003 is used to input the rewritten query information into the trained music search model, and obtain music search results for the target music through the music search model.

[0115] The modules in the aforementioned query information rewriting model training device and music search device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0116] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 11As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores music query information. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When executed by the processor, the computer program implements a query information rewriting model training method or a music search method.

[0117] Those skilled in the art will understand that Figure 11 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0118] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0119] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0120] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0121] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0122] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0123] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0124] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for training a query information rewriting model, characterized in that, The method includes: Obtain sample music query information from sample users, as well as the user preferences of the sample users for sample music search results; wherein, the sample music query information carries emoji information, and the sample music search results are music search results obtained through the sample music query information; The sample music query information and the user preference are input into the query information rewriting model to be trained. The emoji information carried in the sample music query information is rewritten by the query information rewriting model to obtain the predicted query rewriting information corresponding to the sample music query information. By utilizing the difference between the predicted query rewriting information and the actual query rewriting information corresponding to the sample music query information, the query information rewriting model is trained to obtain the trained query information rewriting model.

2. The method according to claim 1, characterized in that, The step of rewriting the emoji information carried in the sample music query information using the query information rewriting model to obtain the predicted query rewriting information corresponding to the sample music query information includes: The emoji information is mapped using the query information rewriting model to obtain initial query rewriting information; The initial query rewrite information is merged with the sample music query information to obtain candidate query rewrite information; Based on the user preferences, the predicted query rewrite information is obtained from the candidate query rewrite information.

3. The method according to claim 2, characterized in that, The step of obtaining the predicted query rewriting information from the candidate query rewriting information based on the user preference includes: Based on the user preferences, the candidate query rewriting information is sorted to obtain the sorting result of the candidate query rewriting information; Based on the sorting results, a preset number of candidate query rewriting information is obtained as the predicted query rewriting information.

4. The method according to claim 1, characterized in that, The acquisition of sample music query information from sample users, and the user preferences of the sample users for sample music search results, includes: Obtain pre-built search behavior logs; The historical music query information used by the sample user to search for sample music, the user interaction results of the sample user with respect to the historical music search results, and the user preference settings of the sample user are obtained from the search behavior log; wherein the historical music search results are music search results obtained through the historical music query information. The historical music query information is used as the sample music query information, and the user interaction results and the user preference settings are used as the user preferences.

5. The method according to claim 4, characterized in that, Before obtaining the pre-built search behavior logs, the process also includes: In response to the sample user's music search request for sample music, obtain the sample user's historical music query information used to search for sample music, and the historical music search results corresponding to the historical music query information; Obtain the user interaction results of the sample users with respect to the historical music search results, as well as the user preference settings of the sample users; The search behavior log is constructed using the historical music query information, the historical music search results, the user interaction results, and the user preference settings.

6. A music search method, characterized in that, The method includes: In response to a target user's music search request for a target music, the system obtains the music query information input by the target user for searching the target music, as well as the target user's first user preference corresponding to historical music search requests; the music query information carries emoji information. The query information rewriting model trained with the music query information and the first user preference input is used to rewrite the emoji information carried in the music query information to obtain query rewritten information corresponding to the music query information; wherein, the query information rewriting model is trained by the query information rewriting model training method as described in any one of claims 1 to 5. The rewritten query information is input into the trained music search model, and music search results for the target music are obtained through the music search model.

7. The method according to claim 6, characterized in that, After obtaining the music search results corresponding to the music query information through the music search model, the process further includes: Obtain the target user's second user preference for the music search results; Based on the music query information and the second user preference, a search behavior log of the target user for the target music is constructed; the search behavior log is used to obtain a new first user preference when a new music search request is received from the target user again; wherein the new first user preference includes the second user preference.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images