Vocabulary understanding method and device, equipment and storage medium

Abstract

The invention discloses a vocabulary understanding method and device, equipment and a storage medium, and relates to the technical field of computers. The method comprises the following steps: acquiring input information; song information corresponding to the input information is obtained from a song information base, the song information corresponding to the input information comprises song information corresponding to at least one vocabulary in the input information, the song information corresponding to the vocabulary comprises total song information corresponding to the vocabulary and converted song information corresponding to the vocabulary, and the vocabulary comprises at least one character; understanding information corresponding to the input information is generated through the trained vocabulary understanding model according to the input information and the song information corresponding to the input information, and the understanding information is used for indicating entity word information contained in the input information and intention information of the input information. The input information is preliminarily understood by obtaining the song information corresponding to the input information, the vocabulary understanding difficulty is reduced, and the vocabulary understanding accuracy is improved by generating the intention information.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a vocabulary understanding method, device, equipment and storage medium. Background Art

[0002] Currently, when people want to listen to music, they usually enter a search term in a music application, and the search engine of the music application obtains corresponding songs according to the search term and plays them.

[0003] In the related technology, the search engine of the music application needs to understand the search terms first, and obtain the entity words and search intent contained in the search terms by adopting the model of extracting entities and intents. The model of extracting entities and the model of intent recognition share the underlying encoding structure of BERT (Bidirectional Encoder Representations from Transformers, a bidirectional language representation model based on the Transformer architecture). After the search terms are segmented, the token representation of each character is obtained, and then each token is marked using the sequence labeling method in the entity recognition layer, and the [CLS] is classified in the intent recognition layer, and finally the entity words and search intent contained in the search terms are output.

[0004] However, the models in the above methods have insufficient understanding of search terms, which easily leads to problems of entity nesting and entity spanning, resulting in inaccurate vocabulary understanding. Summary of the invention

[0005] The embodiments of the present application provide a vocabulary understanding method, apparatus, device and storage medium. The technical solutions provided by the embodiments of the present application are as follows:

[0006] According to one aspect of an embodiment of the present application, a vocabulary understanding method is provided, the method comprising:

[0007] Get input information;

[0008] Acquire song information corresponding to the input information from a song information database, wherein the song information corresponding to the input information includes song information corresponding to at least one word in the input information, wherein the song information corresponding to the word includes full song information corresponding to the word and converted song information corresponding to the word, wherein the full song information corresponding to the word is used to indicate type information and weight information of the word, and the converted song information corresponding to the word is used to indicate historical search information of the word, wherein the word includes at least one character;

[0009] Through the trained vocabulary comprehension model, comprehension information corresponding to the input information is generated according to the input information and the song information corresponding to the input information, and the comprehension information is used to indicate the entity word information contained in the input information and the intention information of the input information.

[0010] According to one aspect of an embodiment of the present application, a vocabulary understanding device is provided, the device comprising:

[0011] An input information acquisition module, used to acquire input information;

[0012] A song information acquisition module, used to acquire song information corresponding to the input information from a song information database, wherein the song information corresponding to the input information includes song information corresponding to at least one word in the input information, wherein the song information corresponding to the word includes full song information corresponding to the word and converted song information corresponding to the word, wherein the full song information corresponding to the word is used to indicate type information and weight information of the word, and the converted song information corresponding to the word is used to indicate historical search information of the word, wherein the word contains at least one character;

[0013] The understanding information generation module is used to generate understanding information corresponding to the input information based on the input information and the song information corresponding to the input information through a trained vocabulary understanding model, wherein the understanding information is used to indicate the entity word information contained in the input information and the intention information of the input information.

[0014] According to one aspect of an embodiment of the present application, a server is provided, the server comprising a processor and a memory, the memory storing a computer program, the computer program being loaded and executed by the processor to implement the above-mentioned vocabulary understanding method.

[0015] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, in which a computer program is stored. The computer program is loaded and executed by a processor to implement the above-mentioned vocabulary understanding method.

[0016] According to one aspect of an embodiment of the present application, a computer program product is provided. The computer program product includes a computer program. The computer program is loaded and executed by a processor to implement the above-mentioned vocabulary understanding method.

[0017] The technical solution provided in the embodiments of the present application can bring the following beneficial effects:

[0018] By obtaining the song information corresponding to the input information, the input information can be preliminarily understood according to the song information corresponding to at least one word in the input information in the song information library, so that when the trained vocabulary understanding model understands the input information according to the input information and the song information corresponding to the input information, the difficulty of vocabulary understanding for the model can be reduced, the vocabulary understanding time of the model can be saved, and the vocabulary understanding efficiency can be improved. And because the understanding information generated by the trained vocabulary understanding model includes the entity word information in the input information and the intention information of the input information, the model can understand the input information more thoroughly, avoid the problems of entity nesting and entity spanning, and improve the accuracy of vocabulary understanding. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of an implementation environment of a solution provided by an embodiment of the present application;

[0020] Figure 2 is a flowchart of a vocabulary understanding method provided by an embodiment of the present application;

[0021] Figure 3 is a schematic diagram of a vocabulary comprehension process provided by an embodiment of the present application;

[0022] Figure 4 is a block diagram of a vocabulary understanding device provided by an embodiment of the present application;

[0023] Figure 5 This is a structural block diagram of a server provided by an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0025] Please refer to Figure 1 , which shows a schematic diagram of a solution implementation environment provided by an embodiment of the present application. The solution implementation environment can be implemented as a vocabulary understanding system. The solution implementation environment may include: a terminal device 10 and a server 20.

[0026] The number of terminal devices 10 can be one or more. The terminal device 10 can be an electronic device such as a mobile phone, a tablet computer, a laptop computer, a desktop computer, a game console, an e-book reader, a multimedia player, a wearable device, an intelligent voice interaction device, a smart home appliance, a vehicle terminal, an aircraft, etc.

[0027] The terminal device 10 may be installed with a client of a music application, and the music application has the function of displaying resources corresponding to the input information according to the information input by the user, and the user can select resources from the resources corresponding to the input information for playback. Optionally, the resources may be audio resources or video resources. Optionally, the music application may be an application that needs to be downloaded and installed, or an application that can be used instantly, and this application does not limit this.

[0028] The server 20 is used to provide background services for the client of the music application installed and running in the terminal device 10. For example, the server 20 can be the background server of the above application. The server 20 can be an independent physical server, or a server cluster composed of multiple servers, or a cloud computing service center. Optionally, the server 20 provides background services for the applications in multiple terminal devices 10 at the same time. The terminal device 10 and the server 20 can communicate with each other through the network.

[0029] In an embodiment of the present application, the server generates resources corresponding to the input information based on the user's input information. In the process of the server generating resources corresponding to the input information based on the input information, the server needs to first perform vocabulary understanding on the input information and obtain understanding information corresponding to the input information, specifically including entity word information contained in the input information and intention information of the input information. Then, the server generates resources corresponding to the input information based on the understanding information corresponding to the input information. Among them, the server obtains song information corresponding to the input information from the song information library based on the input information, and generates understanding information corresponding to the input information based on the input information and the song information corresponding to the input information through the trained vocabulary understanding model.

[0030] Please refer to Figure 2 , which shows a flowchart of a vocabulary understanding method provided by an embodiment of the present application. The execution subject of each step of the method may be a server. The method may include at least one of the following steps 210 to 230:

[0031] Step 210, obtaining input information.

[0032] The input information refers to the information input by the user in the search box of the music application, and the input information includes at least one character. For example, the input information is "pure music of little cute baby".

[0033] Step 220, obtaining song information corresponding to the input information from the song information library, the song information corresponding to the input information includes song information corresponding to at least one word in the input information, the song information corresponding to the word includes full song information corresponding to the word and converted song information corresponding to the word, the full song information corresponding to the word is used to indicate the type information and weight information of the word, the converted song information corresponding to the word is used to indicate the historical search information of the word, and the word contains at least one character.

[0034] The song information library is an information library constructed based on the full domain dictionary and historical input information, wherein the full domain dictionary is a dictionary constructed based on at least one configuration information of at least one song in the music library of the music application, and the historical input information refers to the input information within the historical time period. The song information library includes the full song information corresponding to at least one vocabulary in the full domain dictionary, and the converted song information corresponding to at least one historical input information. The specific creation process of the song information library can refer to the following embodiment, which will not be introduced in detail here.

[0035] The input information includes at least one word, each word includes at least one character in the input information, and when the word includes multiple characters, each word includes multiple consecutive characters in the input information. According to the at least one word included in the input information, song information corresponding to the at least one word is obtained from the song information library.

[0036] If a certain word from at least one of the words is included in the full domain dictionary, or the word is included in the historical input information, it means that the song information corresponding to the word is included in the song information library, and the song information corresponding to the input information includes the song information corresponding to the word. For example, if the full domain dictionary contains the word, the song information corresponding to the word includes the full song information corresponding to the word. If the historical input information contains the word, the song information corresponding to the word includes the converted song information corresponding to the word. If the full domain dictionary does not contain a certain word from at least one of the words, and the historical input information does not contain the word, it means that the song information library does not include the song information corresponding to the word, and the song information corresponding to the input information does not include the song information corresponding to the word.

[0037] Exemplarily, if the input information is "pure music of little cute treasure", at least one word contained in the input information may include: little, cute, treasure, of, pure, sound, music, little cute, cute treasure, treasure's, of pure, pure sound, music, little cute treasure, cute treasure's, treasure's pure, of pure sound, pure music, little cute treasure's, cute treasure's pure, treasure's pure sound, of pure music, little cute treasure's pure, cute treasure's pure sound, treasure's pure music, little cute treasure's pure sound, cute treasure's pure music, little cute treasure's pure sound, cute treasure's pure music, little cute treasure's pure music. If some of the words do not belong to the full domain dictionary or the historical input information, the song information corresponding to the input information does not include the song information corresponding to the part of the words. For example, the song information corresponding to "Pure Music of Little Cute Baby" includes the song information corresponding to "Little", the song information corresponding to "Cute", the song information corresponding to "Bao", the song information corresponding to "De", the song information corresponding to "Pure", the song information corresponding to "Sound", the song information corresponding to "Music", the song information corresponding to "Little Cute", the song information corresponding to "Pure Sound", the song information corresponding to "Music", the song information corresponding to "Little Cute Baby" and the song information corresponding to "Pure Music".

[0038] The full amount of song information corresponding to the vocabulary is used to indicate the type information of the vocabulary and the weight information of the vocabulary. The type information of the vocabulary is used to indicate the type to which the vocabulary belongs. The type to which the vocabulary belongs refers to the vocabulary type to which the configuration information of the song indicated by the vocabulary belongs. The vocabulary type of the configuration information includes at least one of the following: song title, person name, album name, cultural name, and qualifier. The type information of the vocabulary includes at least one of the following: song, person, album, cultural name, and qualifier. Exemplarily, a song can be represented by song, a person can be represented by person, an album can be represented by album, cultural name can be represented by ip, and qualifiers can be represented by filter. The weight information of the vocabulary is used to indicate the total number of times at least one song corresponding to the configuration information of the song indicated by the vocabulary has been played. For the creation process of the full amount of song information, please refer to the following embodiments, which will not be introduced in detail here.

[0039] Optionally, if the type information of a word is one type of information, that is, the word belongs to only one type, the full song information corresponding to the word includes the type information and the weight information of the word of the type information. For example, if the type information of the word is a song title, and the number of times the song corresponding to the song title is played is 100, the weight information of the song title is 100, and the full song information corresponding to the word is {"song", "100"}.

[0040] Optionally, if the type information of a word is multiple type information, that is, the word belongs to multiple types, then the full song information corresponding to the word includes multiple type information and weight information corresponding to the words of the multiple type information. For example, if the type information of the word includes a song title and a person's name, and the number of times the song corresponding to the song title is played is 100, then the weight information of the song title is 100, and the number of times the song corresponding to the person's name is played is 200, then the weight information of the person's name is 200, and the full song information corresponding to the word is {"song","100","person","200"}.

[0041] The converted song information corresponding to the vocabulary is used to indicate the historical search information of the vocabulary. Since the vocabulary is historical input information, the historical search information of the vocabulary is used to indicate the historical song playback of the vocabulary. For example, the historical search information of the vocabulary may include the played songs corresponding to the vocabulary, the number of times the played songs corresponding to the vocabulary are played, and the configuration information of the played songs corresponding to the vocabulary. For the creation process of the converted song information, reference may be made to the following embodiments, which will not be described in detail here.

[0042] Step 230, using the trained vocabulary understanding model, based on the input information and the song information corresponding to the input information, generates understanding information corresponding to the input information, where the understanding information is used to indicate the entity word information contained in the input information and the intention information of the input information.

[0043] The trained vocabulary comprehension model is used to perform vocabulary comprehension on input information, specifically, outputting comprehension information corresponding to the input information based on the input information and song information corresponding to the input information.

[0044] In some embodiments, the entity word information includes entity words and type information of the entity words. The type information of the entity words includes at least one of the following: songs, people, albums, cultural creations, and limited words.

[0045] Entity words refer to words with actual meaning in the input information, and entity words include at least one of the following: song title, singer name, album name, cultural name, and qualifier. For example, if the input information is "Little Cute Baby's Pure Music", the entity words contained in the input information are "Little Cute Baby" and "Pure Music".

[0046] The song title refers to the name of the song, such as the song titles of Qilixiang, Bingyu, and Dadi. The type information of the song title is song, which can be represented by song. The name of a person includes the name of the singer, lyricist, composer, arranger, and other names of the people involved in the production and singing of the song. The type information of the name of a person is person, which can be represented by person. The album name refers to the name of the media in which multiple songs are released together, such as the album names of I am Busy, The Answer is You, and Destiny Party. The type information of the album name is album, which can be represented by album. The name of cultural and creative industries includes the names of literary works, film and television, variety shows, labels, animation, games, and other cultural and creative industries. The type information of the cultural and creative name is cultural and creative, which can be represented by ip. The qualifier refers to the words that restrict or modify the song title, person name, album name, and cultural and creative name, which are used to narrow the search scope, such as the qualifiers such as male version, female version, down-tuned version, ending song, folk song, and first season. The type information of the qualifier is the qualifier word, which can be represented by filter.

[0047] In some embodiments, entity words may also include audio names, video names, etc. Audio names refer to the names of audio resources, including but not limited to novel explanations, movie soundtracks, TV series soundtracks, operas, crosstalk, and other audio resources. Video names refer to the names of video resources, including but not limited to MV (Music Video), concert videos, performance videos, and other video resources. Accordingly, the type information of entity words may also include audio, video, etc. The type information of audio names is audio, which can be represented by radio, and the type information of video names is video, which can be represented by video.

[0048] In some embodiments, when the number of characters included in the input information is less than or equal to the third number, the number of characters included in the entity word is greater than 1.

[0049] The value of the third number is set by the technician according to the vocabulary understanding requirements, and this application does not limit it. For example, the third number is 3, then when the number of characters contained in the input information is less than or equal to 3, the number of characters contained in at least one entity word generated by the model is greater than 1.

[0050] By making the number of characters contained in the entity word greater than 1 when the number of characters contained in the input information is less than or equal to the third number, the complete entity word of the input information is avoided from being split into multiple scattered entities, thereby avoiding the situation where a single character becomes a word, so that the entity words in the input information can be retained to the greatest extent, avoiding inaccurate understanding of the vocabulary.

[0051] In some embodiments, the intent information is used to indicate the resource type pointed to by the input information, and the intent information includes at least one of the following: song, audio, playlist, video.

[0052] Intent information is used to indicate the type of resource the user is looking for. For example, when a user inputs an entity word, the user may want to find the song corresponding to the entity word, or may want to find the playlist corresponding to the entity word, or may want to find the video corresponding to the entity word. The model needs to understand the intent information contained in the input information so that the search engine of the music application can generate resources based on the intent information, and the resources finally displayed on the display interface of the music application can meet the user's expectations.

[0053] A song refers to a specific single sung by a singer, which can be represented by song. If the input information is used to point to a song, a song by a singer, or a song in a TV series, movie, or variety show, the intent information of the input information is song.

[0054] Audio includes but is not limited to novel explanations, movie soundtracks, TV series soundtracks, operas, crosstalk, etc., which can be represented by radio. If the input information is used to point to an audio resource such as a complete or partial novel explanation, the soundtrack of a movie or TV series, a complete collection of operas, a collection of operas by an opera actor, or a complete collection of crosstalk by a crosstalk actor, the intent information of the input information is audio.

[0055] A playlist refers to a collection of songs of a certain type, which can be represented by special. If the input information is not used to indicate a certain song or audio, but uses some suggestive words to indicate the type of song the user wants to listen to, the intent information of the input information is a playlist. The song type includes the style type of the song and the applicable scene type of the song. The style type of the song includes but is not limited to sad, relaxing, hypnotic, cheerful and other styles. The resources pointed to by the input information include but are not limited to sad style playlists, relaxing style playlists, hypnotic style playlists, cheerful style playlists and other styles of playlists. The applicable scene types of songs include but are not limited to commuting scenes, wedding scenes, running scenes and other scene types. The resources pointed to by the input information include but are not limited to songlists suitable for commuting scenes, songlists suitable for wedding scenes, songlists suitable for running scenes and other applicable scene types.

[0056] Video refers to, but is not limited to, MV, concert video, performance video, etc., which can be represented by video. If the input information is used to indicate a video resource, and the intention of watching in the input information is stronger than the intention of listening, then the intention information of the input information is video.

[0057] It should be noted that the input information can contain multiple intents, that is, there are multiple possibilities for intents. In this case, the intent with the highest probability should be output as intent information based on the probabilities of the multiple intents. For example, if the input information contains audio intent and video intent, the intent information is audio if the probability of audio intent is greater than that of video intent, and video if the probability of video intent is greater than that of audio intent.

[0058] By including the intent information of the input information in the understanding information, the model can have a more thorough understanding of the input information. Therefore, when the search engine of the music application generates resources based on the entity word information and intent information, the accuracy of resource generation can be improved, so that the resources finally displayed on the display interface of the music application can better meet the user's expectations.

[0059] For example, if the input information is "Xiao Meng Bao's Pure Music", the understanding information corresponding to the input information is {"entity":{"Xiao Meng Bao":"song","Pure Music":"filter"},"intent":"song"}, where entity represents the entity word information contained in the input information, including the entity word "Xiao Meng Bao" and the type information "song" of "Xiao Meng Bao", as well as the entity word "Pure Music" and the type information "filter" of "Pure Music", "Xiao Meng Bao" is the song name, and "Pure Music" is the qualifier. Intent represents the intent information of the input information, which is the song "song".

[0060] A diagram of the vocabulary comprehension process can be found in Figure 3 As shown, based on the input information, the song information corresponding to the input information is obtained from the song information library, and then based on the input information and the song information corresponding to the input information, the combined information is obtained, and the combined information is input into the trained vocabulary comprehension model to generate comprehension information corresponding to the input information.

[0061] In some embodiments, the training process of the vocabulary understanding model includes the following steps:

[0062] Obtain at least one training sample of the vocabulary understanding model, each training sample includes sample input information and sample understanding information corresponding to the sample input information, the sample understanding information is used to indicate the entity word information contained in the sample input information and the intention information of the sample input information, and the vocabulary understanding model is a pre-trained large language model. Obtain song information corresponding to the sample input information from the song information library. Generate predicted understanding information corresponding to the sample input information through the vocabulary understanding model according to the sample input information and the song information corresponding to the sample input information. Train the vocabulary understanding model with the goal of reducing the difference between the sample understanding information and the predicted understanding information to obtain a trained vocabulary understanding model.

[0063] The vocabulary understanding model is a pre-trained large language model that has the ability to understand the vocabulary of input information. The large language model here can be any publicly available large language model, such as a natural language model based on a transformer structure trained with a large amount of data. The large amount of data can reach a sample level of more than 100 million, and this application does not limit this. It should be noted that the vocabulary understanding ability of the original vocabulary understanding model is weaker than the vocabulary understanding ability of the trained vocabulary understanding model.

[0064] The sample understanding information is the understanding information pre-annotated according to the sample input information, and is used to indicate the entity word information contained in the sample input information and the intent information of the sample input information. The song information corresponding to the sample input information includes the song information corresponding to at least one word in the sample input information, and the song information corresponding to the word includes the full song information corresponding to the word and the converted song information corresponding to the word. The full song information corresponding to the word is used to indicate the type information and weight information of the word, and the converted song information corresponding to the word is used to indicate the historical search information of the word, and the word contains at least one character.

[0065] The sample input information and the song information corresponding to the sample input information are input into the vocabulary understanding model, and the predicted understanding information corresponding to the sample input information is output. The predicted understanding information is used to indicate the entity word information contained in the sample input information predicted by the model and the intent information of the sample input information predicted by the model.

[0066] The loss function value is calculated based on the difference between the characters at each position of the sample understanding information and the predicted understanding information. According to the loss function value, the parameters of the vocabulary understanding model are adjusted to obtain the trained vocabulary understanding model. For the loss function value, any loss function that can represent the meaning of loss can be selected, including but not limited to the mean square error loss function, the cross entropy loss function, the log-likelihood loss function, the exponential loss function, etc., which is not limited in this application.

[0067] By using sample input information and sample understanding information to further adjust model parameters of the pre-trained vocabulary comprehension model, the trained vocabulary comprehension model improves vocabulary comprehension ability and improves the accuracy of vocabulary comprehension. Therefore, applying the trained vocabulary comprehension model to the search engine can improve the accuracy of search results.

[0068] The technical solution provided by the embodiment of the present application can obtain the song information corresponding to the input information, so that the input information can be preliminarily understood according to the song information corresponding to at least one word in the input information in the song information library, so that when the trained vocabulary understanding model understands the input information according to the input information and the song information corresponding to the input information, the difficulty of the model for vocabulary understanding can be reduced, the vocabulary understanding time of the model can be saved, and the vocabulary understanding efficiency can be improved. And because the understanding information generated by the trained vocabulary understanding model includes the entity word information in the input information and the intention information of the input information, the model can understand the input information more thoroughly, avoid the problems of entity nesting and entity spanning, and improve the accuracy of vocabulary understanding. Therefore, when the understanding information of the input information is used to search for the resources corresponding to the input information, the accuracy of the generated resources can be improved, so that the resources finally displayed on the display interface of the music application can be more in line with the user's expectations.

[0069] The following is an introduction to the construction process of the song information database. The method for generating the song information database includes at least one of steps S1 to S3.

[0070] Step S1, based on at least one configuration information of at least one song in the music library, generate full song information corresponding to at least one configuration information, the configuration information of the song includes at least one of the following: song title, person name, album name, cultural name, and qualifier.

[0071] According to at least one configuration information of at least one song in the music library, a full domain dictionary is generated, and the full domain dictionary includes the vocabulary name of at least one configuration information of at least one song in the music library. The configuration information of the song includes at least one of the following: song title, person name, album name, cultural name, and qualifier. Therefore, the full domain dictionary includes the song title of at least one song in the music library, the name of at least one song, the album name of at least one song, the cultural name of at least one song, and the qualifier of at least one song, among which the name of the person includes the name of the singer, the name of the lyricist, the name of the composer, the name of the arranger, and other names of people involved in the production and singing of the song.

[0072] In some embodiments, the same configuration information of different songs may be the same. For example, the names of different songs may be the same, for example, the names of a song A by singer A and a song B by singer B are the same, or singer A and singer B both sing the same song A, generating two different versions of song A. And the names of singers, lyricists, composers, arrangers, and other names of different songs may also be the same, for example, singer A sings songs A and B, or lyricist A writes lyrics for both songs A and B, or composer A composes both songs A and B, and so on. And the album names of different songs may be the same, for example, song A and song B are both in the same album A, and are two songs in album A. And the cultural and creative names of different songs may be the same, for example, song A and song B are both two songs in the same TV series. And the qualifiers of different songs may be the same, for example, the qualifiers of song A and song B are both "slowed down version".

[0073] In some embodiments, different configuration information of different songs may be the same. Exemplarily, the song titles and names of different songs may be the same, for example, the song title of song A and the name of singer B corresponding to song B are the same. And the song titles and album names of different songs may be the same, for example, the song title of song A and the album name of album B where song B is located are the same. And the song titles and cultural names of different songs may be the same, for example, the song title of song A and the cultural name corresponding to song B are the same. And the song titles and qualifiers of different songs may be the same, for example, the song title of song A and the qualifiers of song B are the same. And so on.

[0074] In some embodiments, step S1 includes at least one of sub-steps S11 to S13.

[0075] Sub-step S11, determining type information corresponding to at least one item of configuration information according to at least one item of configuration information of at least one song, where the type information is used to indicate the vocabulary type to which the configuration information belongs.

[0076] The type information of the configuration information includes at least one of the following: song, person, album, cultural creation, and qualifier. The type information is used to indicate the type of vocabulary to which the configuration information belongs, that is, if the configuration information is a song title, the type information of the configuration information is a song, which can be represented by song; if the configuration information is a person name, the type information of the configuration information is a person, which can be represented by person; if the configuration information is an album name, the type information of the configuration information is an album, which can be represented by album; if the configuration information is a cultural creation name, the type information of the configuration information is cultural creation, which can be represented by ip; if the configuration information is a qualifier, the type information of the configuration information is a qualifier, which can be represented by filter.

[0077] According to the configuration information of each song, the configuration information of at least one song in the music library is integrated to obtain at least one configuration information of at least one song. The type information of the configuration information of at least one song is integrated to obtain type information corresponding to at least one configuration information. Optionally, each configuration information may correspond to only one type information or may correspond to multiple type information.

[0078] If the configuration information only corresponds to one type information, it means that the configuration information is only used to represent one configuration of the song in the music library. For example, if the configuration information only corresponds to the type information "song", it means that the configuration information is only used to represent the song title in the music library, and will not be used to represent the name of the song, album name, cultural name, or qualifier. Optionally, this configuration information can be used only to represent one configuration of a song in the music library. For example, if configuration information a only corresponds to the type information "song", configuration information a can be used only to represent the title of song A. Optionally, this configuration information can also be used to represent the same configuration of multiple songs. For example, configuration information a can be used to represent the titles of song A and song B.

[0079] If the configuration information corresponds to multiple types of information, it means that the configuration information is used to represent different configurations of songs in the music library. For example, if the configuration information corresponds to the type information "song" and "personnel", it means that the configuration information is used to represent the song title and person name of the song in the music library, and will not be used to represent the album name, cultural name, and qualifier of the song. Optionally, this configuration information can be used to represent a configuration of a song and another configuration of a song in the music library. For example, if the configuration information a corresponds to the type information "song" and "personnel", the configuration information a can be used to represent the song title of song A and the name of song B. Optionally, this configuration information can be used to represent the same configuration of multiple songs and another configuration of another song in the music library. For example, the configuration information a can be used to represent the song titles of songs A and B, and the name of song C. Optionally, this configuration information can be used to represent a configuration of a song and another configuration of multiple songs in the music library. For example, the configuration information a can be used to represent the song title of song A, and the names of songs B and C. Optionally, this configuration information can be used to represent the same configuration of multiple songs and another configuration of multiple songs in the music library. For example, configuration information a can be used to represent the song titles of song A and song B, and the names of people of song C and song D.

[0080] Sub-step S12, based on at least one item of configuration information of at least one song, obtain weight information corresponding to at least one item of configuration information, where the weight information is used to indicate the total number of times the at least one song corresponding to the configuration information is played.

[0081] The weight information corresponding to each item of configuration information of each song is the same, which is the number of times the song has been played. For example, the weight information corresponding to the title of song A is the number of times song A has been played, and the weight information corresponding to the name of the person, album name, cultural name, and qualifier of song A are also the number of times song A has been played.

[0082] The at least one song corresponding to the configuration information refers to at least one song with the same configuration information and the same type information of the configuration information, including at least one song with the same configuration information. For example, if configuration information a corresponds to a type information "song", then the at least one song corresponding to configuration information a refers to at least one song whose song title is the configuration information a. If configuration information a corresponds to multiple types of information "song" and "personnel", then the at least one song corresponding to configuration information a includes two cases, namely, at least one song when the type information of configuration information a is "song" (the song title is at least one song of configuration information a), and at least one song when the type information of configuration information a is "personnel" (the person name is at least one song of configuration information a).

[0083] The weight information corresponding to the configuration information is used to indicate the total number of times at least one song with the same configuration information and the same type information of the configuration information is played. For example, if configuration information a corresponds to a type information "song", the weight information corresponding to configuration information a is used to indicate the total number of times at least one song with the same song name as configuration information a is played. If configuration information a corresponds to multiple types of information "song" and "personnel", the weight information corresponding to configuration information a is divided into the weight information when the type information of configuration information a is "song" (the total number of times the song name is played for at least one song with the same configuration information a), and the weight information when the type information of configuration information a is "personnel" (the total number of times the song name is played for at least one song with the same configuration information a).

[0084] For each item of configuration information, according to the type information corresponding to the configuration information, determine at least one song with the same configuration information and the same type information of the configuration information, and then according to the number of times each song is played, obtain weight information corresponding to the configuration information with the same configuration information and the same type information of the configuration information, and the weight information corresponding to the configuration information is the sum of the number of times the configuration information and the same type information of the configuration information are played.

[0085] In the case where the configuration information corresponds to a type of information, the weight information of the configuration information is used to indicate the total number of times at least one song with the same configuration information is played. Exemplarily, if the configuration information is a song name, the weight information of the configuration information is used to indicate the total number of times at least one song with the same song name is played. For example, if the configuration information a is the song name of Song A and Song B, the weight information of the configuration information a is the total number of times Song A and Song B are played. If the configuration information is a person's name, the weight information of the configuration information is used to indicate the total number of times at least one song with the same person's name is played. For example, if the configuration information a is the name of Song A and Song B, the weight information of the configuration information a is the total number of times Song A and Song B are played. If the configuration information is an album name, the weight information of the configuration information is used to indicate the total number of times at least one song in the album is played. For example, if the configuration information a is the album name of the album where Song A and Song B are located, the weight information of the configuration information a is the total number of times Song A and Song B are played. If the configuration information is a cultural name, the weight information of the configuration information is used to indicate the total number of times at least one song under the cultural name is played. For example, if the configuration information a is the cultural name of the cultural name of songs A and B, the weight information of the configuration information a is the total number of times songs A and B are played. If the configuration information is a qualifier, the weight information of the configuration information is used to indicate the total number of times at least one song with the same qualifier is played. For example, if the configuration information a is the qualifier of songs A and B, the weight information of the configuration information a is the total number of times songs A and B are played.

[0086] In the case where the configuration information corresponds to multiple types of information, the weight information of the configuration information is used to indicate the total number of times the configuration information and the type information of the configuration information are the same and the total number of times at least one song is played. Exemplarily, if the type information of the configuration information corresponds to songs and people, the weight information of the configuration information is divided into the weight information when the type information of the configuration information is songs (the total number of times the song name is played of at least one song in the configuration information), and the weight information when the type information of the configuration information is people (the total number of times the person name is played of at least one song in the configuration information).

[0087] Exemplarily, if the configuration information is "Little Cute Treasure", and the type information of the configuration information includes songs and albums, then the weight information when the type information of the configuration information is songs is the total number of times at least one song whose song name is "Little Cute Treasure" is played, for example, the weight information when the type information of the configuration information is songs is 103, and the weight information when the type information of the configuration information is albums is the total number of times at least one song whose album name is "Little Cute Treasure" is played, for example, the weight information when the type information of the configuration information is albums is 76.

[0088] Sub-step S13: Generate the full-song information corresponding to each of the at least one piece of configuration information according to the type information corresponding to each of the at least one piece of configuration information and the weight information corresponding to each of the at least one piece of configuration information.

[0089] The full-song information corresponding to the configuration information includes the type information corresponding to the configuration information and the weight information of the type information corresponding to the configuration information. Exemplarily, if the configuration information is "Little Baby", the full-song information corresponding to the configuration information is expressed as {"Little Baby": "song", "103", "album", "76"}.

[0090] Generate the full-song information corresponding to each of the at least one piece of configuration information according to the type information corresponding to each of the at least one piece of configuration information and the total number of plays of at least one song with the same configuration information and the same type information of the configuration information.

[0091] By constructing the full-song information through the above method, the weight information of each vocabulary in the full-domain dictionary under each type information can be obtained. Thus, based on the full-song information, the playing situation of at least one vocabulary in the input information under different type information can be initially obtained, which helps to reduce the difficulty of the model's understanding of vocabulary and saves the model's vocabulary understanding time.

[0092] Exemplarily, if the input information is "Pure music of Little Baby", the full-song information corresponding to at least one vocabulary in the input information can be expressed as {"Pure music": "person", "358787", "Little": "person", "1220", "Pure tone", "album", "305", "song", "82", "Baby", "person", "282", "Music", "album", "184", "Little", "song", "45", "person", "59", "Music", "album", "143", "song", "56", "Baby", "person", "126", "album", "47", "Little Baby", "song", "103", "album", "76", "Pure", "song", "77", "album", "25", "Tone", "album", "77", "song", "35", "Of", "song", "46", "album", "20", "Little", "song", "46", "album", "20"}.

[0093] Step S2: Generate the converted song information corresponding to each of the at least one historical input information according to the playing results of the at least one historical input information, where the playing results of the historical input information are used to indicate the played songs corresponding to the historical input information.

[0094] The historical input information refers to the input information within the historical time period, and the played song corresponding to the historical input information refers to the song that the user finally selects to play after inputting the historical input information.

[0095] In some embodiments, step S2 includes at least one of sub-steps S21 to S23.

[0096] Sub-step S21, for the same historical input information, according to the playback result of the historical input information, obtain the playback times of the played songs, sort the played songs in descending order according to the playback times of the played songs, and obtain the sorted played songs.

[0097] For at least one identical historical input information, the number of times each song in the playback result is played is obtained according to the playback result of at least one historical input information. The number of times a song is played refers to the number of times a user inputs the same historical input information and ultimately chooses to play the same song. The number of times a song is played is used to indicate the number of times at least one song is played when the same historical input information is ultimately played. For example, after the user inputs at least one historical input information, the playback result of at least one historical input information includes song A, song B, and song C. The user chooses to play song A 10 times, i.e., the number of times song A is played is 10. The user chooses to play song B 8 times, i.e., the number of times song B is played is 8. The user chooses to play song C 7 times, i.e., the number of times song C is played is 7.

[0098] At least one of the played songs in the play result is sorted in descending order according to the number of times the played songs are played, to obtain sorted played songs. For example, the number of times song A is played is 10, the number of times song B is played is 8, and the number of times song C is played is 7. Then song A, song B, and song C are sorted in descending order according to the number of times the played songs are played, to obtain sorted played songs, which are song A, song B, and song C in that order.

[0099] Sub-step S22, obtaining the first number of played songs ranked at the top among the sorted played songs.

[0100] The value of the first number is set by the technician according to the vocabulary understanding requirement, and is not limited in this application.

[0101] In some embodiments, if the number of the sorted played songs is greater than the first number, the first number of played songs ranked higher in the sorted played songs are obtained. If the number of the sorted played songs is less than or equal to the first number, the sorted played songs are obtained.

[0102] Sub-step S23, generating conversion song information corresponding to the historical input information according to the configuration information corresponding to the first number of played songs and the number of times the first number of played songs respectively correspond to the first number of played songs.

[0103] The converted song information corresponding to the historical input information includes configuration information of each played song in the first number of played songs and the number of times each song is played.

[0104] Exemplarily, if the historical input information is "how singer A", the converted song information corresponding to the historical input information can be expressed as, {"how singer A":{"number of plays":7156,"song name":"how","singer":"singer A","lyrics":"XXX","composer":"XXX","original singer":"XXX","singer alias":"XXX","song alias":"XXXX"},{"number of plays":6331,"song name":"how","singer":"singer B","album":"XXXX","ip":"XXX","lyrics":"XXX","composer":"XXX","original singer":"XXX","singer alias":"XXX","song alias":"XXX"},{"number of plays":1255,"song name":"how","singer":"singer C","album":"how","lyrics":"XXX","composer":"XXX","original singer":"XXX"},{"number of plays":874,"song name":"how's this","Singer":"XXX","Album":"XXX","Lyrics":"XXX","Composer":"XXX","Translation":"How? X","Singer's Alias":"XXX"},{"Play Count":186,"Song Title":"XXX","Singer":"XXX","Album":"How","Lyrics":"XXX","Composer":"XXX"}.

[0105] By constructing the converted song information through the above method, it is possible to preliminarily obtain the historical search situation of at least one word in the input information based on the converted song information, which helps to reduce the difficulty of the model in understanding the words and save the model's time in understanding the words.

[0106] Step S3, generating a song information library according to the full song information corresponding to at least one configuration information and the converted song information corresponding to at least one historical input information.

[0107] Exemplarily, the song information database is constructed based on RAG (Retrieval-Augmented Generation) technology.

[0108] By generating a song information database, it is possible to obtain song information corresponding to the input information from the song information database, so as to make a preliminary understanding of the input information based on the full song information and the converted song information. When the trained vocabulary comprehension model performs vocabulary comprehension on the input information, it can reduce the difficulty of vocabulary comprehension for the model, save the vocabulary comprehension time of the model, and improve the vocabulary comprehension efficiency.

[0109] Next, the method of generating understanding information is explained.

[0110] In some embodiments, when the input information is preprocessed information, understanding information corresponding to the input information is obtained from the processing result of the preprocessed information, and the processing result of the preprocessed information is the understanding information obtained by pre-comprehension of the preprocessed information by a trained vocabulary comprehension model.

[0111] The preprocessed information refers to input information for vocabulary comprehension by a pre-trained vocabulary comprehension model, and the processing result of the preprocessed information is the comprehension information obtained by pre-comprehension of the preprocessed information by the trained vocabulary comprehension model.

[0112] By directly obtaining the understanding information corresponding to the input information from the processing results of the preprocessing information when the input information belongs to preprocessing information, the time for obtaining the understanding information can be saved, the efficiency of obtaining the understanding information can be improved, and it is avoided that the understanding processing of the vocabulary through the model after obtaining the input information will take too long and affect the user experience.

[0113] In some embodiments, for preprocessing information, at least one historical input information in the first time period that meets the preprocessing condition is obtained every first time period, and the preprocessing condition includes that the historical input information has not been lexically understood by the trained vocabulary understanding model, and the number of occurrences of the historical input information is greater than or equal to the second number. The preprocessing information is updated according to the at least one historical input information in the first time period to obtain updated preprocessing information.

[0114] The value of the first time period is set by the technician according to the requirements of preprocessing, and this application does not limit it. Exemplarily, the first time period can be 1 hour. The value of the second number is set by the technician according to the requirements of vocabulary understanding, and this application does not limit it. Exemplarily, the second number is 3. Then, every hour, historical input information that has not been understood by the trained vocabulary understanding model and has a number of occurrences greater than or equal to 3 is obtained, and the preprocessing information is updated to obtain updated preprocessing information.

[0115] In some embodiments, for the processing result of the preprocessing information, the trained vocabulary understanding model generates understanding information corresponding to the at least one historical input information in the first time period according to the at least one historical input information in the first time period and the song information corresponding to the at least one historical input information in the first time period. According to the understanding information corresponding to the at least one historical input information in the first time period, the processing result of the preprocessing information is updated to obtain the updated processing result of the preprocessing information.

[0116] In some embodiments, since the full song information and the converted song information may be changed within the first time period, after obtaining the updated preprocessing information, the trained vocabulary understanding model generates understanding information corresponding to the updated preprocessing information according to the updated preprocessing information and the song information corresponding to the updated preprocessing information. According to the understanding information corresponding to the updated preprocessing information, the processing result of the preprocessing information is updated to obtain the processing result of the updated preprocessing information.

[0117] After obtaining the updated preprocessing information, it is determined whether the input information belongs to the updated preprocessing information. If the input information belongs to the updated preprocessing information, the understanding information corresponding to the input information in the processing result of the updated preprocessing information is obtained.

[0118] By updating the preprocessing information and the processing result of the preprocessing information every first time period, the existing understanding information can be obtained in time, avoiding the influence of no update or untimely update on vocabulary understanding efficiency and thus the influence on user experience.

[0119] In some embodiments, when the input information does not belong to pre-processed information, the original vocabulary understanding model is used to generate understanding information corresponding to the input information according to the input information and song information corresponding to the input information.

[0120] The original vocabulary understanding model is a vocabulary understanding model that has not been trained in the above-mentioned embodiments. It is a pre-trained large language model. It can be any publicly available large language model, such as a natural language model based on a transformer structure obtained by training with a large amount of data. The large amount of data can reach a sample level of more than 100 million, and this application does not limit this.

[0121] Since the vocabulary comprehension ability of the original vocabulary comprehension model is weaker than that of the trained vocabulary comprehension model, the time required for vocabulary comprehension of the original vocabulary comprehension model is less than the time required for vocabulary comprehension of the trained vocabulary comprehension model.

[0122] By using the original vocabulary comprehension model instead of the trained vocabulary comprehension model to generate comprehension information corresponding to the input information when the input information does not belong to the preprocessed information, the time for vocabulary comprehension can be shortened as much as possible, the efficiency of vocabulary comprehension can be ensured, and the time for vocabulary comprehension can be too long to affect the user experience.

[0123] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0124] Please refer to Figure 4 , which shows a block diagram of a vocabulary understanding device provided by an embodiment of the present application. The device has the function of implementing the above-mentioned vocabulary understanding method, and the function can be implemented by hardware, or by hardware executing corresponding software. The device can be the server introduced above, or it can be set in the server. Figure 4 As shown, the device 400 may include: an input information acquisition module 410, a song information acquisition module 420 and an understanding information generation module 430.

[0125] The input information acquisition module 410 is used to acquire input information.

[0126] The song information acquisition module 420 is used to acquire the song information corresponding to the input information from the song information library, the song information corresponding to the input information includes the song information corresponding to at least one word in the input information, the song information corresponding to the word includes the full song information corresponding to the word and the converted song information corresponding to the word, the full song information corresponding to the word is used to indicate the type information and weight information of the word, the converted song information corresponding to the word is used to indicate the historical search information of the word, and the word contains at least one character.

[0127] The understanding information generation module 430 is used to generate understanding information corresponding to the input information based on the input information and the song information corresponding to the input information through the trained vocabulary understanding model, and the understanding information is used to indicate the entity word information contained in the input information and the intention information of the input information.

[0128] In some embodiments, the apparatus 400 further includes an information base generation module, wherein the information base generation module is configured to:

[0129] Generate full song information corresponding to at least one configuration information of at least one song in the music library, wherein the configuration information of the song includes at least one of the following: song title, person name, album name, cultural name, and qualifier;

[0130] Generate converted song information corresponding to the at least one historical input information according to the playback result of the at least one historical input information, wherein the playback result of the historical input information is used to indicate the playback song corresponding to the historical input information;

[0131] The song information library is generated according to the full song information corresponding to the at least one configuration information and the converted song information corresponding to the at least one historical input information.

[0132] In some embodiments, the information library generation module is used to:

[0133] Determine, according to at least one item of configuration information of the at least one song, type information corresponding to the at least one item of configuration information, wherein the type information is used to indicate the type of vocabulary to which the configuration information belongs;

[0134] According to at least one item of configuration information of the at least one song, obtaining weight information corresponding to the at least one item of configuration information, wherein the weight information is used to indicate the total number of times the at least one song corresponding to the configuration information is played;

[0135] According to the type information corresponding to the at least one item of configuration information and the weight information corresponding to the at least one item of configuration information, the full amount of song information corresponding to the at least one item of configuration information is generated.

[0136] In some embodiments, the information library generation module is used to:

[0137] For the same historical input information, according to the playing result of the historical input information, the playing times of the played songs are obtained, and the played songs are sorted in descending order according to the playing times of the played songs to obtain sorted played songs;

[0138] Obtaining a first number of played songs ranked top among the sorted played songs;

[0139] According to the configuration information corresponding to the first number of played songs and the number of times the first number of played songs respectively correspond to the first number of played songs, the converted song information corresponding to the historical input information is generated.

[0140] In some embodiments, the apparatus 400 further includes a model training module, wherein the model training module is configured to:

[0141] Acquire at least one training sample of the vocabulary understanding model, each of the training samples includes sample input information and sample understanding information corresponding to the sample input information, the sample understanding information is used to indicate entity word information contained in the sample input information and intent information of the sample input information, and the vocabulary understanding model is a pre-trained large language model;

[0142] Acquire song information corresponding to the sample input information from the song information library;

[0143] Generate predicted understanding information corresponding to the sample input information according to the sample input information and the song information corresponding to the sample input information through the vocabulary understanding model;

[0144] The vocabulary comprehension model is trained with the goal of reducing the difference between the sample comprehension information and the predicted comprehension information to obtain the trained vocabulary comprehension model.

[0145] In some embodiments, the understanding information generating module 430 is used to:

[0146] In the case where the input information is preprocessed information, obtaining understanding information corresponding to the input information in a processing result of the preprocessed information, wherein the processing result of the preprocessed information is understanding information obtained by pre-performing vocabulary understanding on the preprocessed information by the trained vocabulary understanding model;

[0147] In the case that the input information does not belong to the preprocessing information, the understanding information corresponding to the input information is generated according to the input information and the song information corresponding to the input information through the original vocabulary understanding model.

[0148] In some embodiments, the understanding information generating module 430 is used to:

[0149] Acquire at least one piece of historical input information in a first time period that satisfies a preprocessing condition every first time period, wherein the preprocessing condition includes that the historical input information has not been subjected to vocabulary comprehension by the trained vocabulary comprehension model, and that the number of occurrences of the historical input information is greater than or equal to a second number;

[0150] Generate, by the trained vocabulary understanding model, understanding information corresponding to at least one historical input information in the first time period according to at least one historical input information in the first time period and song information corresponding to at least one historical input information in the first time period;

[0151] updating the preprocessing information according to at least one historical input information within the first time period to obtain updated preprocessing information;

[0152] The processing result of the preprocessing information is updated according to the understanding information respectively corresponding to at least one historical input information within the first time period to obtain an updated processing result of the preprocessing information.

[0153] In some embodiments, the entity word information includes entity words and type information of the entity words, and the type information of the entity words includes at least one of the following: songs, people, albums, cultural creation, and limited vocabulary;

[0154] When the number of characters included in the input information is less than or equal to the third number, the number of characters included in the entity word is greater than 1.

[0155] In some embodiments, the intent information is used to indicate the resource type pointed to by the input information, and the intent information includes at least one of the following: song, audio, playlist, video.

[0156] It should be noted that the device provided in the above embodiment, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the content structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0157] Please refer to Figure 5 , which shows a structural block diagram of a server 500 provided in one embodiment of the present application. The server 500 may be any electronic device with data calculation, processing and storage functions. The server 500 may be used to implement the vocabulary understanding method provided in the above embodiment.

[0158] Typically, the server 500 includes a processor 501 and a memory 502 .

[0159] The processor 501 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 501 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field Programmable Gate Array), and PLA (Programmable Logic Array). The processor 501 may also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 501 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 501 may also include an AI processor, which is used to process computing operations related to machine learning.

[0160] The memory 502 may include one or more computer-readable storage media, which may be non-transitory. The memory 502 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 502 is used to store a computer program, which is configured to be executed by one or more processors to implement the above-mentioned vocabulary understanding method.

[0161] Those skilled in the art will understand that Figure 5 The structure shown in the figure does not constitute a limitation on the server 500, and the server 500 may include more or less components than those shown in the figure, or combine some components, or adopt a different arrangement of components.

[0162] In an illustrative embodiment, a computer-readable storage medium is also provided, in which a computer program is stored, and when the computer program is executed by a processor of a server, the above-mentioned vocabulary understanding method is implemented. Optionally, the above-mentioned computer-readable storage medium can be a ROM (Read-Only Memory), a RAM (Random Access Memory), a CD-ROM (Compact Disc Read-Only Memory), a magnetic tape, a floppy disk, an optical data storage device, etc.

[0163] In an exemplary embodiment, a computer program product is also provided, the computer program product comprising a computer program, the computer program being stored in a computer-readable storage medium. A processor of a server reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the server executes the above-mentioned vocabulary understanding method.

[0164] It should be understood that the "multiple" mentioned in this article refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. In addition, the step numbers described in this article only illustrate a possible execution sequence between the steps. In some other embodiments, the above steps may not be executed in the order of the numbers, such as two steps with different numbers are executed at the same time, or two steps with different numbers are executed in the opposite order to the diagram. The embodiments of the present application are not limited to this.

[0165] The above description is only an exemplary embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A vocabulary comprehension method, characterized in that: The method comprises: Get input information; Acquire song information corresponding to the input information from a song information database, wherein the song information corresponding to the input information includes song information corresponding to at least one word in the input information, wherein the song information corresponding to the word includes full song information corresponding to the word and converted song information corresponding to the word, wherein the full song information corresponding to the word is used to indicate type information and weight information of the word, and the converted song information corresponding to the word is used to indicate historical search information of the word, wherein the word includes at least one character; Through the trained vocabulary comprehension model, comprehension information corresponding to the input information is generated according to the input information and the song information corresponding to the input information, and the comprehension information is used to indicate the entity word information contained in the input information and the intention information of the input information.

2. The method according to claim 1, characterized in that The method further comprises: Generate full song information corresponding to at least one item of configuration information of at least one song in the music library, wherein the song configuration information includes at least one of the following: song title, person name, album name, cultural name, and qualifier; Generate converted song information corresponding to the at least one historical input information according to the playback result of the at least one historical input information, wherein the playback result of the historical input information is used to indicate the playback song corresponding to the historical input information; The song information library is generated according to the full song information corresponding to the at least one configuration information and the converted song information corresponding to the at least one historical input information.

3. The method according to claim 2, characterized in that The generating, according to at least one item of configuration information of at least one song in the music library, full song information corresponding to the at least one item of configuration information comprises: Determine, according to at least one item of configuration information of the at least one song, type information corresponding to the at least one item of configuration information, wherein the type information is used to indicate the type of vocabulary to which the configuration information belongs; According to at least one item of configuration information of the at least one song, obtaining weight information corresponding to the at least one item of configuration information, wherein the weight information is used to indicate the total number of times the at least one song corresponding to the configuration information is played; According to the type information corresponding to the at least one item of configuration information and the weight information corresponding to the at least one item of configuration information, the full amount of song information corresponding to the at least one item of configuration information is generated.

4. The method according to claim 2, characterized in that: The generating, according to the playback result of at least one historical input information, the converted song information corresponding to the at least one historical input information, comprises: For the same historical input information, according to the playing result of the historical input information, the playing times of the played songs are obtained, and the played songs are sorted in descending order according to the playing times of the played songs to obtain sorted played songs; Obtaining a first number of played songs ranked top among the sorted played songs; According to the configuration information corresponding to the first number of played songs and the number of times the first number of played songs respectively correspond to the first number of played songs, the converted song information corresponding to the historical input information is generated.

5. The method according to claim 1, characterized in that The method further comprises: Acquire at least one training sample of the vocabulary understanding model, each of the training samples includes sample input information and sample understanding information corresponding to the sample input information, the sample understanding information is used to indicate entity word information contained in the sample input information and intent information of the sample input information, and the vocabulary understanding model is a pre-trained large language model; Acquire song information corresponding to the sample input information from the song information library; Generate predicted understanding information corresponding to the sample input information according to the sample input information and the song information corresponding to the sample input information through the vocabulary understanding model; The vocabulary comprehension model is trained with the goal of reducing the difference between the sample comprehension information and the predicted comprehension information to obtain the trained vocabulary comprehension model.

6. The method according to claim 1, characterized in that The trained vocabulary understanding model generates understanding information corresponding to the input information according to the input information and the song information corresponding to the input information, including: In the case where the input information is preprocessed information, obtaining understanding information corresponding to the input information in a processing result of the preprocessed information, wherein the processing result of the preprocessed information is understanding information obtained by pre-performing vocabulary understanding on the preprocessed information by the trained vocabulary understanding model; In the case that the input information does not belong to the preprocessing information, the understanding information corresponding to the input information is generated according to the input information and the song information corresponding to the input information through the original vocabulary understanding model.

7. The method according to claim 6, characterized in that The method further comprises: Acquire at least one piece of historical input information in a first time period that satisfies a preprocessing condition every first time period, wherein the preprocessing condition includes that the historical input information has not been subjected to vocabulary comprehension by the trained vocabulary comprehension model, and that the number of occurrences of the historical input information is greater than or equal to a second number; Generate, by the trained vocabulary understanding model, understanding information corresponding to at least one historical input information in the first time period according to at least one historical input information in the first time period and song information corresponding to at least one historical input information in the first time period; updating the preprocessing information according to at least one historical input information within the first time period to obtain updated preprocessing information; The processing result of the preprocessing information is updated according to the understanding information respectively corresponding to at least one historical input information within the first time period to obtain an updated processing result of the preprocessing information.

8. The method according to any one of claims 1 to 7, characterized in that: The entity word information includes entity words and type information of the entity words, and the type information of the entity words includes at least one of the following: songs, people, albums, cultural creation, and limited vocabulary; When the number of characters included in the input information is less than or equal to the third number, the number of characters included in the entity word is greater than 1.

9. The method according to any one of claims 1 to 7, characterized in that: The intent information is used to indicate the resource type pointed to by the input information, and the intent information includes at least one of the following: song, audio, playlist, video.

10. A vocabulary comprehension device, characterized in that: The device comprises: An input information acquisition module, used to acquire input information; A song information acquisition module, used to acquire song information corresponding to the input information from a song information database, wherein the song information corresponding to the input information includes song information corresponding to at least one word in the input information, wherein the song information corresponding to the word includes full song information corresponding to the word and converted song information corresponding to the word, wherein the full song information corresponding to the word is used to indicate type information and weight information of the word, and the converted song information corresponding to the word is used to indicate historical search information of the word, wherein the word contains at least one character; The understanding information generation module is used to generate understanding information corresponding to the input information based on the input information and the song information corresponding to the input information through a trained vocabulary understanding model, wherein the understanding information is used to indicate the entity word information contained in the input information and the intention information of the input information.

11. A server, characterized in that: The server includes a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the vocabulary understanding method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which is loaded and executed by a processor to implement the vocabulary understanding method according to any one of claims 1 to 9.

13. A computer program product, characterized in that The computer program product comprises a computer program, and the computer program is loaded and executed by a processor to implement the vocabulary understanding method according to any one of claims 1 to 9.

Images