Music object searching method and device, equipment, storage medium and program product

Abstract

The invention relates to a music object searching method and device, equipment, a storage medium and a program product. The method comprises the steps of performing intention recognition on search information input for a music library, and obtaining at least one predicted intention category corresponding to the search information and respective intention probability values of the predicted intention categories; for each prediction intention category in the at least one prediction intention category, obtaining a historical click probability value of the search information under each prediction intention category; determining an intention quota value of each prediction intention category according to the intention probability value and the historical click probability value of each prediction intention category; and determining an object recall number corresponding to the intention quota value of each predicted intention category, and recalling music objects of the object recall number corresponding to each predicted intention category from a music library. By adopting the method, the search accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a music object search method, apparatus, device, storage medium, and program product. Background Art

[0002] Intent recognition analyzes and categorizes user search input to retrieve the desired search results. It's a crucial component of search. Correct intent recognition can yield satisfying results, while incorrect intent recognition often fails to deliver the desired results. In the music industry, music libraries are vast. In addition to songs released by music companies or artists, they also include user-uploaded data, much of which is simply to capitalize on popularity. Therefore, accurately identifying user search intent is crucial.

[0003] In traditional technology, the intent of the search information input by the user is usually identified based on the intent recognition model, the intent category to which the search information belongs is obtained, and the corresponding data is directly recalled based on the intent category. However, the data recalled in this way often cannot recall the results that the user really wants, resulting in low search accuracy. Summary of the Invention

[0004] Based on this, it is necessary to provide a music object search method, device, equipment, storage medium and program product that can improve search accuracy in response to the above technical problems.

[0005] In a first aspect, the present application provides a music object search method, the method comprising:

[0006] Performing intent recognition on search information inputted into the music library to obtain at least one predicted intent category corresponding to the search information and an intent probability value for each of the predicted intent categories;

[0007] For each predicted intent category in the at least one predicted intent category, obtaining a historical click probability value of the search information under each predicted intent category;

[0008] Determining the intent quota value of each predicted intent category according to the intent probability value of each predicted intent category and the historical click probability value of each predicted intent category;

[0009] Determine the object recall quantity corresponding to the intention quota value of each predicted intention category, and recall the music objects of the object recall quantity corresponding to each predicted intention category from the music library.

[0010] In a second aspect, the present application provides a music object search device, the device comprising:

[0011] an identification module for performing intent recognition on search information input for a music library, and obtaining at least one predicted intent category corresponding to the search information and an intent probability value for each of the predicted intent categories;

[0012] an acquisition module, configured to obtain, for each predicted intent category in the at least one predicted intent category, a historical click probability value of the search information under each predicted intent category; and determine an intent quota value for each predicted intent category based on the intent probability value of each predicted intent category and the historical click probability value of each predicted intent category;

[0013] The recall module is used to determine the object recall quantity corresponding to the intention quota value of each predicted intention category, and recall the music objects of the object recall quantity corresponding to each predicted intention category from the music library.

[0014] In a third aspect, the present application provides a computer device including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps in the method embodiments of the present application are implemented.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the various method embodiments of the present application.

[0016] In a fifth aspect, the present application provides a computer program product, including a computer program, which implements the steps in the various method embodiments of the present application when the computer program is executed by a processor.

[0017] The above-mentioned music object search method, device, equipment, storage medium and program product, by performing intent recognition on the search information input for the music library, obtains at least one predicted intent category corresponding to the search information and the intent probability value of each predicted intent category; for each predicted intent category in at least one predicted intent category, obtains the historical click probability value of the search information under each predicted intent category; determines the intent quota value of each predicted intent category based on the intent probability value of each predicted intent category and the historical click probability value of each predicted intent category; determines the number of object recalls corresponding to the intent quota value of each predicted intent category, and recalls the music objects corresponding to the object recall number of each predicted intent category from the music library. Compared with the traditional music object search method, the present application adjusts the intent probability value of the predicted intent category obtained by intent recognition through the historical click probability value, that is, adjusts the intent probability value according to the search needs of the actual user group, obtains the intent quota value, and then recalls the corresponding music objects from the music library based on the intent quota value, which can recall the results that the user really wants, thereby improving the search accuracy of the music object. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A diagram illustrating an application environment of a music object search method according to an embodiment;

[0019] Figure 2 1 is a flow chart of a music object search method according to an embodiment;

[0020] Figure 3 A schematic diagram of the principle of counting historical click probability values ​​of each predicted intent category in one embodiment;

[0021] Figure 4 A schematic flow chart of a music object search method according to another embodiment;

[0022] Figure 5 is a structural block diagram of a music object search device in one embodiment;

[0023] Figure 6 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0025] The music object search method provided by this application can be applied to Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can be separately configured to store data that server 104 needs to process. The data storage system can be integrated with server 104 or placed in the cloud or on another server. Terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, in-vehicle terminals, intelligent voice interaction devices, aircraft, smart home appliances, and portable wearable devices. Smart home appliances can include smart speakers, smart TVs, and smart air conditioners. Portable wearable devices can include smart watches, smart bracelets, head-mounted devices, etc. Server 104 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides network security services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, cloud security, host security, CDN, and basic cloud computing services such as big data and artificial intelligence platforms. Terminal 102 and server 104 can be connected directly or indirectly via wired or wireless communication, which is not limited in this application.

[0026] The server 104 may perform intent recognition on the search information input for the music library, and obtain at least one predicted intent category corresponding to the search information and the intent probability value of each predicted intent category. For each predicted intent category in at least one predicted intent category, the server 104 may obtain the historical click probability value of the search information under each predicted intent category. The server 104 may determine the intent quota value of each predicted intent category based on the intent probability value of each predicted intent category and the historical click probability value of each predicted intent category. The server 104 may determine the number of object recalls corresponding to the intent quota value of each predicted intent category, and recall the music objects corresponding to the object recall number of each predicted intent category from the music library. The server 104 may send the recalled music objects to the terminal 102 and display them in the terminal 102.

[0027] It is understandable that Figure 1 The application scenarios are only for illustration and the present application is not limited thereto.

[0028] In one embodiment, Figure 2 As shown, a music object search method is provided. This embodiment takes the method applied to the server 104 as an example for description, and includes the following steps:

[0029] In step 202, the search information inputted into the music library is subjected to intent recognition to obtain at least one predicted intent category corresponding to the search information and the intent probability value of each predicted intent category.

[0030] The search information is information input by the user for searching for a music object (song) from a music library, and the search information may be a search text input by the user.

[0031] In one embodiment, after a user enters search information for a music library, the server performs intent recognition on the search information to obtain at least one predicted intent category and an intent probability value for each predicted intent category, where the predicted intent category may be included in a predicted intent category list. It is understood that the predicted intent category may include at least one of a song (name), singer (name), album (name), film (name), lyrics, or semantics. The intent probability value of a predicted intent category is the probability value that the predicted intent category is the intent category expected by the user.

[0032] In one embodiment, the server may input the search information into the intent recognition model, which performs intent recognition on the search information and outputs a predicted intent category list corresponding to the search information, which contains at least one predicted intent category, and the intent probability values ​​of each predicted intent category contained in the predicted intent category list.

[0033] For example, the server can input the search information "Fairy Tale of Fish and Butterflies" into the intent recognition model. The list of predicted intent categories identified by the intent recognition model includes two predicted intent categories: "song" and "lyrics." The intent probability value of the predicted intent category "song" is 0.8, and the intent probability value of the predicted intent category "lyrics" is 0.2. It can be understood that when a user enters the search information "Fairy Tale of Fish and Butterflies", the probability that this search information is the title of a song is 0.8, and the probability that it is a line of lyrics is 0.2.

[0034] Step 204 : For each predicted intent category in at least one predicted intent category, obtain a historical click probability value of the search information under each predicted intent category.

[0035] Among them, the music library contains multiple music objects, each of which corresponds to multiple intent categories. To distinguish it from the above-mentioned predicted intent categories, the intent categories corresponding to the music objects can be called library preset intent categories, that is, each music object corresponds to multiple library preset intent categories. A library preset intent category is used to describe a music object from one dimension, so multiple different library preset intent categories of a music object are used to describe the music object from multiple different dimensions. Among the multiple library preset intent categories corresponding to each music object, there is at least one predicted intent category obtained by intent recognition of the search information.

[0036] The historical click probability value for a search information under a preset music library intent category refers to the ratio of the number of clicks (category clicks) by users on that music library intent category for the same search information to the total number of clicks by users across all preset music library intent categories. Therefore, it can be understood that the historical click probability value for each predicted intent category refers to the ratio of the number of clicks (category clicks) by users on that predicted intent category for the same search information to the total number of clicks by users across all preset music library intent categories. It should be noted that the user is any user from the past, not the user currently entering the search information. Furthermore, the number of clicks by users on music library intent categories, specifically the number of clicks by users on music objects within that preset music library intent category, indicates that a user's click on a music object within a particular preset music library intent category indicates a higher interest in that music object within that category and that the music object within that category better meets the user's search expectations. Overall, the historical click probability value for each search information under each preset music library intent category can reflect the proportion of users' desired intents for each search information within the overall user population.

[0037] Specifically, the server can count the historical click data for the search information, which includes the record data of users clicking on the intent category of the search information. Based on these record data, the historical click probability value of the search information under any preset intent category of the music library can be determined.

[0038] Taking the search information "Fairy Tale of Fish and Butterflies" as an example, its historical click data may include a record data (user A1 clicked on a music object recalled by the search information "Fairy Tale of Fish and Butterflies" at time T1), a record data (user A2 clicked on a music object recalled by the search information "Fairy Tale of Fish and Butterflies" at time T1), and a record data (user A3 clicked on a music object recalled by the search information "Fairy Tale of Fish and Butterflies" at time T2).

[0039] For the music library preset intent category "song" among the multiple music library preset intent categories corresponding to the music object, when the predicted intent category obtained by intent recognition of the search information also includes the predicted intent category "song", then the historical click probability value of the search information "Fairy Tale of Fish and Butterflies" under the music library preset intent category "song" can be determined based on these recorded data.

[0040] Step 206 : Determine the intent quota value of each predicted intent category based on the intent probability value of each predicted intent category and the historical click probability value of each predicted intent category.

[0041] Among them, the intent quota value of the predicted intent category is the ratio of the number of music objects belonging to the predicted intent category recalled when recalling music objects from the music library to the total number of recalled music objects.

[0042] In one embodiment, determining the intent quota value of the predicted intent category based on the intent probability value and the historical click probability value of the predicted intent category includes: performing a weighted summation on the intent probability value and the historical click probability value of the predicted intent category to obtain the intent quota value of the predicted intent category. Specifically, the server may perform a weighted summation on the intent probability value and the historical click probability value of the predicted intent category to obtain a weighted value, and determine the intent quota value of the predicted intent category based on the weighted value.

[0043] In one embodiment, the server may perform a weighted summation of the intent probability value and the historical click probability value of the predicted intent category to obtain a weighted value, and directly use the weighted value as the intent quota value of the predicted intent category.

[0044] In one embodiment, the intent quota value of the predicted intent category can be calculated using the following formula:

[0045] ;

[0046] Among them, x represents the predicted intent category, represents the intent probability value of the predicted intent category, It represents the historical click probability value of the predicted intent category, 0.3 and 0.7 are the pre-set weight values ​​of the intent probability value and the historical click probability value, respectively. It is understood that the weight values ​​of the intent probability value and the historical click probability value can also be set according to actual needs, and this embodiment does not limit this.

[0047] In one embodiment, the server may select one of the intent probability value and the historical click probability value of the predicted intent category as the intent quota value of the predicted intent category. For example, the one with the larger value may be selected as the intent quota value of the predicted intent category.

[0048] Step 208: Determine the object recall quantity corresponding to the intent quota value of each predicted intent category, and recall the music objects of the object recall quantity corresponding to each predicted intent category from the music library.

[0049] Specifically, the server may recall the number of music objects indicated by the intent quota value for each predicted intent category from the music library. Furthermore, the server may send the recalled music objects to the terminal corresponding to the user who entered the search information, so that the recalled music objects can be displayed to the user via the terminal.

[0050] In one embodiment, the server may determine a preset total number of recalls and obtain the number of objects to be recalled corresponding to the intent quota value of each predicted intent category based on the product of the intent quota value of each predicted intent category and the total number of recalls. For example, if the intent quota value is 0.3 and the total number of recalls is 10, the number of objects to be recalled is 3.

[0051] In the above-mentioned music object search method, by performing intent recognition on the search information input for the music library, at least one predicted intent category corresponding to the search information and the intent probability value of each predicted intent category are obtained; for each predicted intent category in at least one predicted intent category, the historical click probability value of the search information under each predicted intent category is obtained; based on the intent probability value of each predicted intent category and the historical click probability value of each predicted intent category, the intent quota value of each predicted intent category is determined; the number of object recalls corresponding to the intent quota value of each predicted intent category is determined, and music objects corresponding to the number of object recalls of each predicted intent category are recalled from the music library. Compared with the traditional music object search method, the present application adjusts the intent probability value of the predicted intent category obtained by intent recognition through the historical click probability value, that is, adjusts the intent probability value according to the search needs of the actual user group, obtains the intent quota value, and then recalls the corresponding music objects from the music library based on the intent quota value, which can recall the results that the user really wants, thereby improving the search accuracy of the music object.

[0052] In one embodiment, the method also includes: obtaining multiple music library preset intention categories corresponding to each music object in the music library; wherein each music object corresponds to multiple music library preset intention categories; multiple music library preset intention categories include at least one predicted intention category; determining the intention category included in the multiple music library preset intention categories but not included in at least one predicted intention category as a cold start intention category, and obtaining the intention features corresponding to each cold start intention category; if the intention features corresponding to the cold start intention category meet the preset category addition conditions, determining the initial intention probability value and each initial historical click probability value of the cold start intention category; determining the intention quota value of each cold start intention category based on the initial intention probability value and initial historical click probability value of each cold start intention category; determining the object recall quantity corresponding to the intention quota value of each cold start intention category, and recalling the music objects corresponding to the cold start intention category from the music library.

[0053] The intent feature corresponding to the cold start intent category is feature information of historical interaction data corresponding to the cold start intent category.

[0054] It is understandable that a situation in which the predicted intent categories are A, B, and D, while the multiple preset intent categories corresponding to each music object in the music library are A, B, and C, is impossible. This is because the intent recognition model used to identify the intent of search information input from the music library is trained based on the multiple preset intent categories corresponding to each music object in the music library. Therefore, the predicted intent category obtained by performing intent recognition on the input search information using the trained intent recognition model will definitely be included in the multiple preset intent categories corresponding to each music object in the music library.

[0055] In one embodiment, the predicted intent category list includes at least one predicted intent category obtained by performing intent recognition on the search information input for the music library. The music library preset intent category list includes a plurality of music library preset intent categories corresponding to each music object in the music library. When the cold start intent category is not included in the predicted intent category list, the server can obtain the intent features corresponding to each music library preset intent category included in the music library preset intent category list, and search for the intent features corresponding to the cold start intent category from the intent features corresponding to each music library preset intent category. The server can determine whether the intent features corresponding to the cold start intent category meet the preset category addition conditions. When the intent features corresponding to the cold start intent category meet the preset category addition conditions, the server can determine the initial intent probability value and the initial historical click probability value of the cold start intent category. It can be understood that the initial intent probability value and the initial historical click probability value of the cold start intent category are pre-set fixed values. For example, the initial intent probability value is 0.2, and the initial historical click probability value is also 0.2. The server can determine the intent quota value of each cold start intent category based on the initial intent probability value and initial historical click probability value of the cold start intent category, and recall the music objects corresponding to the cold start intent category from the music library based on the object recall quantity corresponding to the intent quota value of each cold start intent category.

[0056] For example, the intent feature corresponding to the "song" intent category might include the total number of plays of all songs with the same name as search information. The intent feature corresponding to the "song" intent category then satisfies the pre-set category addition condition, specifically, that the total number of plays of all songs with the same name as search information exceeds a pre-set threshold, such as 50.

[0057] In one embodiment, after recalling the number of music objects indicated by each cold start intention category from the music library, the method further includes: sorting the music objects recalled from the music library according to the initial intention probability value of each cold start intention category.

[0058] In the above embodiment, when the cold start intent category is not included in the predicted intent category list, it means that the cold start intent category has not been identified through the intent recognition process, and the user search cannot recall the music objects belonging to the cold start intent category, and there is no corresponding user click data for the cold start intent category. At this time, the intent feature corresponding to the cold start intent category is used to determine whether to add a new cold start intent category to recall the corresponding music object. After determining the new cold start intent category, the cold start intent category is assigned an initial intent probability value and an initial historical click probability value to recall the music object indicated by the cold start intent category from the music library, solving the cold start problem of the new intent category, thereby further improving the search accuracy of the music object.

[0059] In one embodiment, the intent features corresponding to each cold start intention category are obtained, including: obtaining historical interaction data of search information under each cold start intention category; and counting the historical interaction data under each cold start intention category to construct the intent features corresponding to each cold start intention category.

[0060] The historical interaction data of the search information under the cold start intent category is the interaction data of each historical user for the search information under the cold start intent category before the current search.

[0061] Specifically, the server can obtain historical interaction data for search information under the cold start intent category and collect statistics on the historical interaction data to construct the intent feature corresponding to the cold start intent category. It is understandable that in this way, the intent feature corresponding to each preset intent category of the music library included in the reference category list can be constructed.

[0062] In one embodiment, the intent categories may include songs, singers, albums, movies, and lyrics. For the intent category of "songs", the server may obtain the historical interaction data of the search information under the intent category of "songs", and count the historical interaction data to construct the intent features corresponding to the intent category of "songs". For example, the intent features corresponding to the intent category of "songs" may include: the total number of plays of all songs whose names are search information, the number of plays of the most popular songs among all songs whose names are search information, the highest singer level among all singers whose names are search information, the maximum number of song comments among all songs whose names are search information, and the maximum number of song collections among all songs whose names are search information. The intent features corresponding to the intent category of "singers" may include: the total number of plays of all singers whose names are search information, the total number of plays of the most popular singer whose names are search information, the number of plays of the most popular song among all songs whose names are search information, the highest singer level among all singers whose names are search information, and the maximum number of singer fans among all singers whose names are search information. The intent features corresponding to the "Album / Movie" intent category may include: the total number of plays of all songs in the album / movie whose name is the search information, the total number of plays of the most popular songs in the album / movie whose name is the search information, and the number of plays of the most popular songs in the album / movie whose name is the search information. The intent features corresponding to the "Lyrics" intent category may include: the total number of plays of all songs that contain the search information as lyrics, the number of plays of the most popular songs that contain the search information as lyrics, the total number of comments on all songs that contain the search information as lyrics, the number of comments on the most popular songs that contain the search information as lyrics, the total number of collections of all songs that contain the search information as lyrics, and the number of collections of the most popular songs that contain the search information as lyrics.

[0063] In the above embodiment, historical interaction data for search information under the cold start intent category is obtained and statistically analyzed to construct intent features corresponding to the cold start intent category. This improves the accuracy of the intent features corresponding to the preset intent categories of each music library, thereby further improving the search accuracy of music objects.

[0064] In one embodiment, obtaining the historical click probability value of the search information under each predicted intent category includes: determining at least one historical music object associated with the search information; wherein the historical music object is a music object that has been clicked in history among all music objects recalled from the music library based on the search information; determining the total number of clicks on the search information and the number of category clicks on the search information under each predicted intent category based on the historical click data of at least one historical music object; and determining the historical click probability value of the search information under the predicted intent category based on the ratio of the category clicks under the predicted intent category to the total number of clicks for each predicted intent category.

[0065] In one embodiment, determining the total number of clicks on the search information based on historical click data of at least one historical music object includes: taking the total number of times the at least one historical music object was clicked as the total number of clicks on the search information.

[0066] For example, suppose that the preset intent categories of the music library in the music library include: songs, singers, albums, and lyrics. Suppose that the predicted intent categories of the search information "Qilixiang" are: songs, albums, and lyrics. Suppose that the following table is the historical click data of the search information "Qilixiang", then the server can determine the total number of clicks on the search information "Qilixiang" based on the number of historical click data of each historical music object indicated by each music object identifier below. It can be understood that each click of a user corresponds to one piece of historical click data. For example, if the historical click data of the historical music object indicated by 0010 are generated by clicks of user A and user B respectively, there will be 2 pieces of historical click data of the historical music object indicated by 0010.

[0067]

[0068] In one embodiment, the server may determine the music objects that have been clicked on in the history of all music objects recalled from the music library based on the search information, and obtain at least one historical music object associated with the search information. The server may determine the total number of clicks on the search information under each music library preset intent category based on the historical click data of at least one historical music object. For each predicted intent category in each music library preset intent category included in the music library preset intent category list, the server may count the historical click data to determine the number of category clicks on the search information under the predicted intent category. Furthermore, the server may determine the historical click probability value of the search information under the predicted intent category based on the ratio of the category click count to the total click count. It is understandable that in this way, the historical click probability values ​​corresponding to the search information under each predicted intent category included in the music library preset intent category list can be obtained. The music library preset intent category list contains multiple music library preset intent categories corresponding to each music object in the music library.

[0069] In one embodiment, the server may directly use the ratio of the number of category clicks to the total number of clicks as the historical click probability value of the search information under the predicted intent category.

[0070] In one embodiment, Figure 3As shown, each music library preset intent category included in the music library preset intent category list includes songs, singers, albums, movies, lyrics and semantics. Specifically, the server can count the historical click data for the search information to determine the total number of clicks on the search information under each music library preset intent category. For the intent category of "song" in each music library preset intent category included in the music library preset intent category list, the server can count the historical click data to determine the number of category clicks on the search information under the intent category of "song". For the intent category of "singer" in each music library preset intent category included in the music library preset intent category list, the server can count the historical click data to determine the number of category clicks on the search information under the intent category of "singer". For the intent category of "album" in each music library preset intent category included in the music library preset intent category list, the server can count the historical click data to determine the number of category clicks on the search information under the intent category of "album". For the "Movie and TV" intent category within each of the preset intent categories in the music library's preset intent category list, the server may count historical click data to determine the number of category clicks on the search information within the "Movie and TV" intent category. For the "Lyrics" intent category within each of the preset intent categories in the music library's preset intent category list, the server may count historical click data to determine the number of category clicks on the search information within the "Lyrics" intent category. For the "Semantics" intent category within each of the preset intent categories in the music library's preset intent category list, the server may count historical click data to determine the number of category clicks on the search information within the "Semantics" intent category. It is understood that if the search information does not match any of the intent categories (song, artist, album, movie and TV, or lyrics), "Semantics" will be used as the intent category for the search information. Furthermore, for each music library's preset intent category, the server may determine the intended click ratio (i.e., the historical click probability value) for the search information within the predicted intent category based on the ratio of the number of category clicks for the predicted intent category to the total number of clicks.

[0071] In the above embodiment, the total number of clicks on the search information and the number of category clicks on the search information for each predicted intent category are determined based on the historical click data of at least one historical music object associated with the search information. For each predicted intent category, the historical click probability value of the search information for that predicted intent category is determined based on the ratio of the category click count to the total click count. This improves the accuracy of the historical click probability values ​​corresponding to each predicted intent category of the search information, thereby further improving the search accuracy of music objects.

[0072] In one embodiment, each click of each historical music object corresponds to a piece of historical click data, and the historical click data contains the intent content of the historical music object corresponding to the historical click data under each predicted intent category; the number of category clicks of the search information under each predicted intent category is determined based on the historical click data of at least one historical music object, including: for each piece of historical click data, according to the intent matching strategy corresponding to each predicted intent category, the search information is intent-matched with at least one intent content contained in the historical click data to determine the predicted intent category to which the historical click data is successfully matched; for each predicted intent category, the number of historical click data that successfully matches the predicted intent category is determined as the number of category clicks of the search information under the predicted intent category.

[0073] To better understand the process of intent matching between search information (e.g., "Fairy Tale of Fish and Butterflies") and various intent contents in historical click data, let's take an example. The historical click data contains the intent contents of music objects corresponding to each predicted intent category (e.g., song and singer) (e.g., the song title is "Fairy Tale of Fish and Butterflies" and the singer name is "Zhang San"). If the search information "Fairy Tale of Fish and Butterflies" successfully matches the song title "Fairy Tale of Fish and Butterflies", then the historical click data is deemed to match the predicted intent category "song".

[0074] In the above embodiment, the search information is matched with the intent content of each of the multiple historical click data to determine the number of historical click data that match the predicted intent category. The number of historical click data that match the predicted intent category is then used as the category click count for the search information under the predicted intent category. This improves the accuracy of obtaining the category click count corresponding to each predicted intent category for the search information, thereby further improving the search accuracy for music objects.

[0075] In one embodiment, the intent matching strategy includes an exact matching mode, a sequential matching mode, and a complete matching mode; each predicted intent category corresponds to at least one of the exact matching mode, the sequential matching mode, and the complete matching mode; the exact matching mode has a higher matching priority than the sequential matching mode, and the sequential matching mode has a higher matching priority than the complete matching mode; when a matching mode with a higher matching priority is successfully matched, the matching mode with a lower matching priority is skipped; for each piece of historical click data, respectively, according to the intent matching strategy corresponding to each predicted intent category, the search information is intent-matched with at least one intent content contained in the historical click data to determine the predicted intent category for which the historical click data is successfully matched, including: for each piece of historical click data, respectively, according to the matching priority of each matching mode corresponding to each predicted intent category, and the matching priority corresponding to each predicted intent category, the search information is intent-matched with at least one intent content contained in the historical click data to determine the predicted intent category for which the historical click data is successfully matched.

[0076] It can be understood that the exact matching mode indicates that the search information is completely consistent with the intent content to be matched. For example: the search information is "Fairy Tale of Fish and Butterflies", and the intent content to be matched = "Fairy Tale of Fish and Butterflies", then it is considered that the search information is successfully matched with the intent content to be matched. The sequential matching mode indicates that the search information is part of the intent content to be matched. For example: the search information is "Fish and Butterflies", and the intent content to be matched is "Fairy Tale of Fish and Butterflies", then it is considered that the search information is successfully matched with the intent content to be matched. The complete matching mode indicates that although the search information is not part of the intent content to be matched, each word in the search information can be hit sequentially in the intent content to be matched. For example: the search information is "Fish and Butterflies", and the intent content to be matched is "Fairy Tale of Fish and Butterflies", then it is considered that the search information is successfully matched with the intent content to be matched.

[0077] In one embodiment, in the historical click data for search information, there may be multiple cases where the intent content is the same. For example, the song name and lyrics are the same, and the song name and album name are the same. At this time, the matching must follow the following rules: For the intent category of "song": the song name has the highest matching priority, and the exact matching mode, sequential matching mode and complete matching mode are allowed. For the intent category of "singer": the singer name uses the exact matching mode, or only the singer name matches the search information among all intent contents, and it is a sequential matching mode. For the intent category of "album / movie": the exact matching mode, sequential matching mode and complete matching mode are allowed, but when the search information has matched the song name, the album / movie match is ignored. For the intent category of "lyrics": when there is no singer, song, album / movie match at the same matching priority, lyrics are allowed to match, and the exact matching mode, sequential matching mode and complete matching mode are allowed.

[0078] In the above embodiment, the intent matching strategy is limited to include exact matching mode, sequential matching mode, and complete matching mode. The exact matching mode has a higher matching priority than the sequential matching mode, and the sequential matching mode has a higher matching priority than the complete matching mode. According to the matching priority of each matching mode corresponding to each predicted intent category, and the matching priority corresponding to each predicted intent category, the search information is matched with at least one intent content contained in the historical click data. In this way, the matching accuracy of the preset intent categories of each music library can be improved, thereby further improving the search accuracy of music objects.

[0079] In one embodiment, Figure 4As shown, the server can perform intent recognition on the search information input for the music library, obtain a list of predicted intent categories corresponding to the search information, and the intent probability values ​​of each predicted intent category included in the predicted intent category list. For each music library preset intent category in the music library preset intent category list corresponding to the music library, when the music library preset intent category is included in the predicted intent category list, the historical click data of the search information is obtained. The historical click data is counted to determine the total number of clicks on the search information under each music library preset intent category. Among them, each music library preset intent category is an intent category included in the music library preset intent category list. For each music library preset intent category in each music library preset intent category, the search information is matched with the intent content corresponding to each music library preset intent category according to the intent matching strategy corresponding to the music library. When the search information matches the intent content corresponding to the predicted intent category, the historical click data is counted to determine the number of category clicks on the search information under the predicted intent category. Based on the ratio of the number of category clicks to the total number of clicks, the historical click probability value of the search information under the predicted intent category is determined. According to the intent probability value and historical click probability value of the predicted intent category, the intent quota value of the predicted intent category is determined, and according to the object recall number corresponding to the intent quota value of the predicted intent category, the music objects of the object recall number indicated by the predicted intent category are recalled from the music library, and the music objects recalled from the music library are sorted according to the intent probability value of each predicted intent category. When the library preset intent category is not included in the list of predicted intent categories, the intent category included in multiple library preset intent categories but not included in at least one predicted intent category is determined as a cold start intent category, the historical interaction data of the search information under the cold start intent category is obtained, and the historical interaction data is counted to construct the intent feature corresponding to the cold start intent category. When the intent feature corresponding to the cold start intent category meets the preset category addition condition, the initial intent probability value and initial historical click probability value of the cold start intent category are determined. According to the initial intent probability value and the initial historical click probability value of the cold start intent category, the intent quota value of each cold start intent category is determined, and according to the object recall quantity corresponding to the intent quota value of each cold start intent category, the music objects of the object recall quantity indicated by the cold start intent category are recalled from the music library, and according to the initial intent probability value of each cold start intent category, the music objects recalled from the music library are sorted. It is understandable that the server can construct a target intent category list based on the intent categories contained in the preset intent category list of the music library, the intent probability values ​​and the historical click probability values ​​corresponding to the intent categories, or the initial intent probability values ​​and the initial historical click probability values ​​corresponding to the intent categories, and normalize the intent probability values, the historical click probability values, the initial intent probability values ​​and the initial historical click probability values ​​in the target intent category list to obtain a normalized target intent category list.Furthermore, the server can recall the music objects corresponding to the search information based on the normalized target intent category list.

[0080] It should be understood that, although the various steps in the flow chart of the above-mentioned embodiments are shown in sequence, these steps are not necessarily performed in sequence. Unless clearly stated herein, the execution of these steps does not have strict order restrictions, and these steps can be performed in other sequences. Moreover, at least a portion of the steps in the above-mentioned embodiments may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0081] In one embodiment, Figure 5 As shown, a music object search device 500 is provided, which specifically includes:

[0082] Identification module 502, configured to perform intent recognition on search information input for a music library, and obtain at least one predicted intent category corresponding to the search information and an intent probability value for each predicted intent category;

[0083] An acquisition module 504 is configured to acquire, for each predicted intent category in the at least one predicted intent category, a historical click probability value of the search information under each predicted intent category; and determine an intent quota value for each predicted intent category based on the intent probability value and the historical click probability value of each predicted intent category.

[0084] The recall module 506 is used to determine the object recall quantity corresponding to the intention quota value of each predicted intention category, and recall the music objects of the object recall quantity corresponding to each predicted intention category from the music library.

[0085] In one embodiment, the recall module 506 is also used to obtain multiple music library preset intention categories corresponding to each music object in the music library; each music object corresponds to multiple music library preset intention categories; multiple music library preset intention categories include at least one predicted intention category; the intention category included in the multiple music library preset intention categories but not included in at least one predicted intention category is determined as a cold start intention category, and the intention feature corresponding to each cold start intention category is obtained; if the intention feature corresponding to the cold start intention category meets the preset category addition condition, the initial intention probability value and the initial historical click probability value of the cold start intention category are determined; the initial intention probability value and the initial historical click probability value of each cold start intention category are determined, and the intention quota value of each cold start intention category is determined; according to the object recall quantity corresponding to the intention quota value of each cold start intention category, the music objects corresponding to the object recall quantity of each cold start intention category are recalled from the music library.

[0086] In one embodiment, the recall module 506 is further used to obtain historical interaction data of the search information under each cold start intention category; and to count the historical interaction data under each cold start intention category to obtain the intention features corresponding to each cold start intention category.

[0087] In one embodiment, the recall module 506 is further configured to sort the music objects recalled from the music library according to the initial intent probability value of each cold start intent category.

[0088] In one embodiment, the acquisition module 504 is also used to determine at least one historical music object associated with the search information; wherein the historical music object is a music object that has been clicked in the history of all music objects recalled from the music library based on the search information; based on the historical click data of at least one historical music object, the total number of clicks on the search information and the number of category clicks on the search information under each predicted intent category are determined; for each predicted intent category, the historical click probability value of the search information under the predicted intent category is determined based on the ratio of the category clicks under the predicted intent category to the total number of clicks.

[0089] In one embodiment, each click of each historical music object corresponds to a piece of historical click data, and the historical click data contains the intent content of the historical music object corresponding to the historical click data under each predicted intent category; the acquisition module 504 is also used to, for each piece of historical click data, respectively, according to the intent matching strategy corresponding to each predicted intent category, intent match the search information with at least one intent content contained in the historical click data to determine the predicted intent category to which the historical click data is successfully matched; for each predicted intent category, the number of historical click data that successfully matches the predicted intent category is determined as the number of category clicks of the search information under the predicted intent category.

[0090] In one embodiment, the intent matching strategy includes an exact matching mode, a sequential matching mode, and a complete matching mode; each predicted intent category corresponds to at least one of the exact matching mode, the sequential matching mode, and the complete matching mode; the matching priority of the exact matching mode is higher than that of the sequential matching mode, and the matching priority of the sequential matching mode is higher than that of the complete matching mode; when a matching mode with a higher matching priority is successfully matched, a matching mode with a lower matching priority is skipped; the acquisition module 504 is also used to perform intent matching on the search information with at least one intent content contained in the historical click data, respectively, for each piece of historical click data, according to the matching priorities of each matching mode corresponding to each predicted intent category, and the matching priorities corresponding to each predicted intent category, so as to determine the predicted intent category for which the historical click data is successfully matched.

[0091] In one embodiment, the acquisition module 504 is further configured to perform weighted summation on the intent probability value of each predicted intent category and the historical click probability value of each predicted intent category to obtain the intent quota value of the predicted intent category.

[0092] In one embodiment, the recall module 506 is further configured to sort the music objects recalled from the music library according to the intent probability value of each predicted intent category.

[0093] The above-mentioned music object search device, by performing intent recognition on the search information input for the music library, obtains at least one predicted intent category corresponding to the search information and the intent probability value of each predicted intent category; for each predicted intent category in at least one predicted intent category, obtains the historical click probability value of the search information under each predicted intent category; determines the intent quota value of each predicted intent category based on the intent probability value of each predicted intent category and the historical click probability value of each predicted intent category; determines the number of object recalls corresponding to the intent quota value of each predicted intent category, and recalls the music objects corresponding to the object recall number of each predicted intent category from the music library. Compared with the traditional music object search method, the present application adjusts the intent probability value of the predicted intent category obtained by intent recognition through the historical click probability value, that is, adjusts the intent probability value according to the search needs of the actual user group, obtains the intent quota value, and then recalls the corresponding music objects from the music library based on the intent quota value, which can recall the results that the user really wants, thereby improving the search accuracy of the music object.

[0094] Each module in the aforementioned music object search device may be implemented in whole or in part through software, hardware, or a combination thereof. Each module may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0095] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a music object search method is implemented.

[0096] It will be understood by those skilled in the art that Figure 6 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.

[0097] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0098] In one embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the steps in the above-mentioned method embodiments when executed by a processor.

[0099] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0100] 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, stored data, displayed data, 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 relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions.

[0101] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. 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 above-described method embodiments. Any reference to memory, storage, database, or other media used in the embodiments provided herein may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0102] 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.

[0103] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A music object search method, characterized in that: The method comprises: Performing intent recognition on search information input into the music library to obtain at least one predicted intent category corresponding to the search information and an intent probability value of each predicted intent category; For each predicted intent category in the at least one predicted intent category, obtaining a historical click probability value of the search information under each predicted intent category; Determining the intent quota value of each predicted intent category according to the intent probability value of each predicted intent category and the historical click probability value of each predicted intent category; Determine the object recall quantity corresponding to the intention quota value of each predicted intention category, and recall the music objects of the object recall quantity corresponding to each predicted intention category from the music library.

2. The method according to claim 1, characterized in that The obtaining of the historical click probability value of the search information under each of the predicted intent categories includes: Determine at least one historical music object associated with the search information; wherein the historical music object is a music object that has been clicked in history among all music objects recalled from the music library based on the search information; Determine the total number of clicks on the search information and the number of category clicks on the search information under each of the predicted intent categories according to the historical click data of the at least one historical music object; For each of the predicted intent categories, the historical click probability value of the search information under the predicted intent category is determined according to the ratio of the category click count under the predicted intent category to the total click count.

3. The method according to claim 2, characterized in that Each click of each of the historical music objects corresponds to a piece of the historical click data, and the historical click data includes the intention content of the historical music object corresponding to the historical click data under each of the predicted intention categories; Determining the number of category clicks of the search information under each of the predicted intent categories according to the historical click data of the at least one historical music object includes: For each piece of the historical click data, respectively, according to the intent matching strategies corresponding to each of the predicted intent categories, respectively, the search information is intent matched with at least one intent content contained in the historical click data to determine the predicted intent category that the historical click data successfully matches; For each of the predicted intent categories, the number of historical click data that successfully matches the predicted intent category is determined as the number of category clicks of the search information under the predicted intent category.

4. The method according to claim 3, characterized in that The intent matching strategy includes an exact matching mode, a sequential matching mode and a complete matching mode; each of the predicted intent categories corresponds to at least one of the exact matching mode, the sequential matching mode and the complete matching mode; the exact matching mode has a higher matching priority than the sequential matching mode, and the sequential matching mode has a higher matching priority than the complete matching mode; when a matching mode with a higher matching priority is successfully matched, a matching mode with a lower matching priority is skipped; The method of performing intent matching on each piece of the historical click data and at least one intent content included in the historical click data according to the intent matching strategy corresponding to each predicted intent category to determine the predicted intent category that the historical click data successfully matches includes: For each piece of historical click data, respectively, according to the matching priorities of the matching modes corresponding to each predicted intent category and the matching priorities corresponding to each predicted intent category, the search information is intent-matched with at least one intent content contained in the historical click data to determine the predicted intent category that the historical click data successfully matches.

5. The method according to claim 1, characterized in that The determining the intent quota value of each predicted intent category according to the intent probability value of each predicted intent category and the historical click probability value of each predicted intent category includes: The intention probability value of each predicted intention category and the historical click probability value of each predicted intention category are weightedly summed to obtain the intention quota value of each predicted intention category.

6. The method according to claim 1, characterized in that After recalling the object recall quantity of music objects corresponding to each of the predicted intention categories from the music library, the method further includes: The music objects recalled from the music library are sorted according to the intention probability values ​​of the predicted intention categories.

7. The method according to claim 1, characterized in that The method further comprises: Obtaining a plurality of music library preset intention categories corresponding to each music object in the music library, wherein each of the music objects corresponds to the plurality of music library preset intention categories; the plurality of music library preset intention categories include the at least one predicted intention category; Determine an intention category included in the plurality of music library preset intention categories but not included in the at least one predicted intention category as a cold start intention category, and obtain an intention feature corresponding to each of the cold start intention categories; If the intention feature corresponding to the cold start intention category meets the preset category addition condition, determine the initial intention probability value and the initial historical click probability value of the cold start intention category; Determining the intent quota value of each cold start intent category according to the initial intent probability value of each cold start intent category and each initial historical click probability value; Determine the object recall quantity corresponding to the intention quota value of each cold start intention category, and recall the music objects of the object recall quantity corresponding to each cold start intention category from the music library.

8. The method according to claim 7, characterized in that The acquiring the intention feature corresponding to each of the cold start intention categories includes: Obtaining historical interaction data of the search information under each of the cold start intention categories; The historical interaction data under each of the cold start intention categories is counted to obtain the intention features corresponding to each of the cold start intention categories.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images