Label processing method and device, electronic equipment and storage medium

Abstract

The invention relates to a label processing method and device, electronic equipment and a storage medium, and relates to the technical field of data processing. Comprising the steps of determining a first target label for describing an application program; taking the text information of the application program as input information of a chat program to obtain a second target label used for describing the application program; and recommending the application program and / or an associated application program related to the application program according to the first target tag and the second target tag corresponding to the query word input by the user. By using the label processing method provided by the invention, accurate application programs and associated application programs can be recommended to the user.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to a label processing method, device, electronic device and storage medium. Background Art

[0002] Currently, after entering a query word in the search box in the app store, the download entrance and introduction of the searched application will pop up at the first level of the app store interface, and related applications related to the query word will pop up at subsequent levels of the first level.

[0003] In the related art, after a query word is input, although the application searched by the query word can be displayed in the first rank of the application store, the related applications displayed in the subsequent rank of the first rank have a low correlation with the application in the first rank. Summary of the invention

[0004] In order to overcome the problems existing in the related art, the present disclosure provides a label processing method, device, electronic device and storage medium.

[0005] According to a first aspect of an embodiment of the present disclosure, a tag processing method is provided, including:

[0006] determining a first target tag for describing the application;

[0007] Using the text information of the application as input information of the chat program to obtain a second target tag for describing the application;

[0008] The application and / or associated applications related to the application are recommended according to the first target tag and the second target tag corresponding to the query word input by the user.

[0009] Optionally, determining a first target tag for describing the application includes:

[0010] The graphic and text information of the application is used as input information of a graphic and text recognition model to obtain the first target label, wherein the graphic and text information is used to describe the application in the form of pictures and texts.

[0011] Optionally, the using the image and text information of the application as input information of an image and text recognition model to obtain the first target label includes:

[0012] Using the image and text information of the application as input information of the image and text recognition model to obtain a first label and the application;

[0013] Cluster multiple applications and divide them into multiple application groups;

[0014] For each of the application groups, a first target tag matching the application group is determined from a plurality of the first tags, where the first target tag is used to describe the application in the application group.

[0015] Optionally, the method further comprises:

[0016] When the number of applications is less than a first preset number or the number of applications is greater than a second preset number, the first target tag is screened out; the number of applications is the number of applications in the application group that matches the first target tag.

[0017] Optionally, the method further comprises:

[0018] Target applications whose distance from a center point of the application is less than a preset distance are screened out from the multiple applications in the application group to form a target application group, wherein the first target tag is used to describe the applications in the target application group.

[0019] Optionally, the using the text information of the application as input information of the chat program to obtain a second target tag for describing the application includes:

[0020] Using the text information of the application as input information of the chat program to obtain a second label;

[0021] For each application group, a second target tag matching the application group is determined from a plurality of second tags, where the second target tag is used to describe the applications in the application group.

[0022] Optionally, the method further comprises:

[0023] The first target label and the second target label are displayed.

[0024] According to a second aspect of an embodiment of the present disclosure, there is provided a label processing device, including:

[0025] A first target tag determining module, configured to determine a first target tag for describing an application;

[0026] A second target tag determination module is configured to use the text information of the application as input information of the chat program to obtain a second target tag for describing the application;

[0027] The recommendation module is configured to recommend the application and / or associated applications related to the application according to the first target tag and the second target tag.

[0028] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, including:

[0029] processor;

[0030] a memory for storing processor-executable instructions;

[0031] Wherein, the processor is configured to:

[0032] Execute the steps of the label processing method provided in the first aspect of the embodiment of the present disclosure.

[0033] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which computer program instructions are stored. When the program instructions are executed by a processor, the steps of the label processing method provided in the first aspect of the present disclosure are implemented.

[0034] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects:

[0035] On the first aspect, providing a first target tag and a second target tag provides more tags than providing the first target tag. A larger number of tags will make the tags more refined, and more accurate applications and related applications can be recommended using more refined tags. Since the recommended applications and related applications are applications that are strongly related to the user's query terms, the recommended applications and related applications can meet user needs and increase the download rate of applications and related applications in the application store.

[0036] Secondly, the number of applications mapped to the first target tag and the second target tag is limited to a reasonable range, thereby making the number of applications recommended by the application store and the number of associated applications more reasonable.

[0037] Thirdly, the chat program can automatically generate new second target tags for applications without the need for staff to manually conceive and generate new second target tags for applications, thereby realizing the automation of tag generation; and the second target tags generated by the chat program can be more suitable for the application, so the recommended applications and related applications will also be more accurate.

[0038] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0040] Figure 1 It is a flowchart of a label processing method shown according to an exemplary embodiment.

[0041] Figure 2 It is an input representation diagram of a BERT model shown according to an exemplary embodiment.

[0042] Figure 3 It is a schematic diagram of a BERT model performing a classification task shown according to an exemplary embodiment.

[0043] Figure 4 It is a schematic diagram of a BERT model performing similarity calculation shown according to an exemplary embodiment.

[0044] Figure 5 It is a schematic diagram of a BERT model performing sequence labeling shown according to an exemplary embodiment.

[0045] Figure 6 It is a plurality of games in different rankings shown on the application store interface after a user inputs "Game A" according to an exemplary embodiment.

[0046] Figure 7 It is a plurality of games in different rankings shown on the application store interface after a user inputs "Magic" according to an exemplary embodiment.

[0047] Figure 8 It is a schematic diagram of graphic and text information input into an optical character recognition model according to an exemplary embodiment.

[0048] Fig. 9 It is a schematic diagram of a Chinese recognition model obtaining a label vector and an application program vector according to an exemplary embodiment.

[0049] Fig.10 It is a schematic diagram of a plurality of different application program groups according to an exemplary embodiment.

[0050] Fig.11 It is a block diagram of a label processing device shown according to an exemplary embodiment.

[0051] Fig.12 It is a block diagram of a label processing device shown according to an exemplary embodiment.

[0052] Fig.13 It is a block diagram of a label processing device shown according to an exemplary embodiment. Detailed implementation manners

[0053] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0054] It should be noted that all actions of obtaining signals, information, or data in the present disclosure are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where the location is located and with the authorization given by the owner of the corresponding device.

[0055] Currently, when a user enters a query term in the search bar of an app store, the label corresponding to the query term will be determined according to the first mapping relationship between the query term and the label; then, the app introduction related to the app mapped by the label and the download entry of the app will be provided to the user. Taking games as an example, when a user enters a game name in the app store, the label corresponding to the game name will be determined, and then the game introduction related to the game mapped by the label and the download entry of the game will be provided to the user. And the determination of game labels and the determination of the first mapping relationship between the query term and the game label can be obtained through the following solutions:

[0056] (1) Use the TF-IDF (term frequency–inverse document frequency) model to establish game labels.

[0057] The game introduction can be input into the TF-IDF model. The TF-IDF model performs word segmentation, removes invalid words such as "de", "di", and "de", and obtains effective keywords; then, the number of occurrences of the keywords in the game introduction is counted, and the keywords with the number of occurrences greater than the preset number are used as the game labels of the game.

[0058] Specifically, the TF-IDF model will obtain game labels through the following steps.

[0059] The first step is to calculate the number of occurrences of the keyword in the game introduction, and this number of occurrences is also called the word frequency.

[0060]

[0061] The second step is to calculate the inverse document frequency.

[0062]

[0063] The inverse document frequency indicates that the more common a keyword is, the larger the denominator of the inverse document frequency is, and the smaller and closer to 0 the inverse document frequency is, indicating that the keyword is not important for distinguishing different game descriptions. For example, the keyword "multiplayer game" appears in multiple game descriptions, and the inverse document frequency corresponding to the keyword "multiplayer game" is small, so the keyword "multiplayer game" cannot distinguish multiple games and cannot form a signature word for a certain game. In the formula for calculating the inverse document frequency, the denominator is added by 1 to avoid the denominator being 0.

[0064] The third step is to calculate TF-IDF.

[0065] TF-IDF = term frequency (TF) × inverse document frequency (IDF) (3)

[0066] TF-IDF is directly proportional to the number of times a keyword appears in a game description, and inversely proportional to the number of times the keyword appears in the entire corpus. Therefore, the TF-IDF value of each keyword in the game description can be calculated, and then arranged in descending order, with the top N keywords used as game tags.

[0067] However, the TF-IDF model has the following disadvantages:

[0068] Disadvantage 1: When the number of keywords is small, the number of game tags obtained will be small. For example, if the number of keywords obtained is 1, then the game tag is also 1. When the number of game tags is small, the accuracy of the games recommended based on game tags is relatively low. Disadvantage 2: TF-IDF only considers two factors: word frequency and inverse document frequency. It cannot understand that a keyword may have different meanings in different sentences. The TF-IDF model cannot capture the semantic differences of the same word in different sentences. It does not consider the contextual information of a word in different sentences. For example, in the two sentences "I'm going to the bank to deposit money" and "I'm going to sleep in a corridor called bank", the bank in the first sentence represents a financial institution, and the bank in the second sentence is a place name. Then the TF-IDF model cannot capture the meaning of the word "bank" in two different sentences. On this basis, the TF-IDF model will treat keywords with different meanings as the same keyword to count the word frequency and inverse document frequency, which may cause the game tags obtained after sorting to fail to accurately describe the game. Naturally, the games recommended to users based on inaccurate game tags cannot meet user needs.

[0069] (2) Use the BM25 (Best Matching 25) model to obtain the first mapping relationship between the query term and the game tag.

[0070] The BM25 model can perform morpheme analysis on the query word to generate multiple morphemes; then calculate the correlation between the keywords in the game introduction and the multiple morphemes; then perform weighted summation of the correlation scores between the multiple morphemes and the keywords to obtain the correlation score between the query word and the keyword; finally, arrange the multiple correlation scores between the query word and the keyword in descending order, and use the keyword with the largest correlation score with the query word as the game tag corresponding to the query word, thereby establishing the first mapping relationship between the query word and the game tag.

[0071] Specifically, the BM25 model calculates the relevance score between the query term and the game description obtained based on the query term search.

[0072]

[0073] Among them, Q represents the query word, q i represents a morpheme after the query word is parsed; d represents a game introduction obtained by the search; W i Represents the morpheme q i The weight of R(q i , d) represents the correlation score between morpheme qi and game introduction d.

[0074] For the weight W in the above formula (4) i It can be defined by the following formula:

[0075]

[0076] Where N is the number of game profiles found; n(q i ) is the number of game descriptions that contain a certain morpheme. According to the definition of IDF, for a given number of game descriptions, the more game descriptions that contain a certain morpheme, the lower the weight of the morpheme. In other words, when many game descriptions contain a certain morpheme, the discrimination of the morpheme is not high, so the importance of using the morpheme to judge the relevance score is relatively low.

[0077] For the correlation score R(q i , d), can be defined by the following formula:

[0078]

[0079] Among them, k1 and k2 are adjustment factors, and k1 is usually set to 2 based on experience; f i is the frequency of occurrence of the morpheme in the game description, qf i is the frequency of occurrence of the morpheme in the query. In most cases, the frequency of occurrence of the morpheme in the query is once, so qf i=1, so the above formula (6) can be simplified to:

[0080]

[0081]

[0082] Among them, b is the adjustment factor, which is usually set to b=0.75 based on experience; dl is the length of the game introduction, and avgdl is the average length of all game introductions.

[0083] From the definition of K, we can see that the adjustment factor b is used to adjust the effect of the length of the game introduction on the relevance score. The larger the adjustment factor b is, the greater the effect on the relevance score is, and vice versa. The longer the game introduction is, the larger the K value will be, and the smaller the relevance score between the morpheme and the game introduction will be. It can be understood that when the game introduction is longer, the larger the K value is, the more morphemes q i The greater the chance, the greater the frequency of occurrence of the morpheme in the game introduction of different lengths qf i The same, longer game introduction and morpheme q i The correlation score between the short game description and the morpheme q i The correlation score between them is small. Substitute K in the above formula (7) into R (q i , d) can be obtained as follows, which is used to fully express the relevance score between morpheme qi and game introduction d:

[0084]

[0085] However, the BM25 model has the following disadvantages:

[0086] The BM25 model is essentially the same as the TF-IDF model. Both of them establish game tags based on the frequency of occurrence of morphemes or keywords, and establish the first mapping relationship between game tags and query terms. However, the BM25 model also does not take into account the contextual information of the same morpheme or keyword in different sentences, so the recommended games are also inaccurate.

[0087] (3) Use the Word2Vec (cbow+skip-gram+hierarchical softmax+Negativesampling) model to determine whether the manually set game tags are accurate.

[0088] After training, the Word2Vec model can obtain the word vectors in the game introduction and the word vectors of the query words; calculate the similarity between the word vectors of the query words and the word vectors in the game introduction; use the vocabulary corresponding to the word vector that achieves the maximum similarity with the query words as the game label; then determine whether the calculated game label is consistent with the manually set game label. If not, it means that the manually set game label is inaccurate.

[0089] The Word2Vec model has two modes:

[0090] The CBOW (Continuous Bag-of-Words Model) model predicts the current word through the context, which is equivalent to deducting a word from a sentence and then predicting the word based on the context of the word.

[0091] The Skip-gram (Continuous Skip-gram Model) mode predicts the context through the current word, which is equivalent to providing a word and then predicting what words will appear before and after the word.

[0092] The Word2Vec model can take into account the contextual meaning of each word in the game description, so the game corresponding to the query word will be more accurate. However, since the word vectors of the query word and the keywords in the game description are one-to-one, when a keyword in the game description is polysemous, there will be similarity calculation errors, which will eventually lead to inaccurate games recommended to users. For example, when the query word is "Xiaomi", two "Xiaomi" keywords appear in the game description. The first Xiaomi word represents mobile phones, and the second Xiaomi word represents food. The query word calculated by the Word2Vec model has the same similarity with these two keywords. In the end, both words may be used as game tags, and it is impossible to determine whether the manually set game tags are accurate.

[0093] (4) DSSM (Deep Structured Semantic Model) is also used to determine whether the manually set game tags are accurate.

[0094] The DSSM model determines the first mapping relationship between query terms and game descriptions in the user's historical search data, maps the query terms and the corresponding game descriptions to the same semantic space through a neural learning network, and trains the DSSM model by maximizing the cosine similarity between the query terms and the game descriptions.

[0095] The DSSM model calculates the cosine similarity between the query word and the keywords in the game introduction. The cosine similarity represents the semantic similarity score between the query word and the vocabulary in the game introduction. The keyword with the maximum semantic similarity score with the query word is then used as the game tag. Finally, the game tag predicted by the DSSM model is checked to see if it is consistent with the manually set game tag. If not, it means that the manually set game tag is inaccurate.

[0096] Although the DSSM model can distinguish polysemous words by calculating the semantic similarity score between the query words and the words in the game description, it has high requirements on the data quality of the input query words and the game description when training the DSSM model. In addition, the DSSM model can only distinguish polysemous words under a single category. For example, when the training data of the DSSM model is the game category, the DSSM model can only distinguish polysemous words under the game category, but cannot distinguish polysemous words under the music category or the video category, and its generalization is low.

[0097] (5) Use the BERT (Bidirectional Encoder Representation from Transformers) model to create game tags.

[0098] The BERT model is explained in several parts below:

[0099] The first part is the input representation of the BERT model. Figure 2 As shown in the figure, if the input is two sentences "my dog ​​is cute" and "he likes playing", three mappings will be performed on each token, namely, word mapping, position mapping and sentence mapping.

[0100] For word mapping (Token Embeddings), a special marker [CLS] is added at the beginning of the first sentence, a [SEP] is added after cute to indicate the end of the first sentence, and a [SEP] is added after ##ing. Here, "playing" is divided into two markers, "play" and "##ing", to achieve word mapping.

[0101] For position embeddings, different words are distinguished by position, and a position is mapped into a low-density vector. For example, "[CLS]my dog ​​is cute[sep]he likes play##ing[SEP]" is mapped into "E0~E 10 ”.

[0102] For the mapping of sentences (Segment Embeddings), different sentences are distinguished by position. The sentence markers of the same sentence are shared, and the sentence markers of different sentences are different. For example, "my dog ​​is cute" is mapped to "E A ”, mapping “he likes playing” to “E B ”.

[0103] In addition, for the BERT model, the input of the BERT model is usually a fixed sentence length, for example, 128. If the input of the BERT model is less than the sentence length, the sentence will be padded to meet the fixed sentence length; if the input of the BERT model exceeds the sentence length, the excess words will be truncated to ensure that the input of the BERT model is a fixed sentence length.

[0104] The second part is the pre-trained language model (Mask LM). The BERT model considers the contextual information of words through the MASK model. The BERT model uses the MASK language model instead of the ordinary language model. It masks a word in the sentence and lets the BERT model guess the possible masked words. For example, 15% of the words in the sentence will be randomly masked and the BERT model will be asked to guess these masked words, so that the BERT model needs to consider the contextual information when encoding a word.

[0105] But there is a problem. When training the MASK language model, special words will be masked. However, when further training the MASK language model, special words will not be masked, which will cause the problem of inconsistent masked words. Therefore, in the BERT model, if a word is in the 15% of the selected words, it will be randomly executed as follows:

[0106] Method 1: Replace with [MASK] with 80% probability, for example, my dog ​​is hairy changes to my dog ​​is [MASK].

[0107] Method 2: There is a 10% probability of replacing it with a random word, such as my dog ​​is hairy becomes my dog ​​is apple.

[0108] Method 3: There is a 10% probability of replacing it with itself, such as my dog ​​is hairy becomes my dog ​​ishairy.

[0109] The purpose of this design is that the BERT model does not know which word the MASK language model replaces. Any word may be replaced. For example, the word "apple" in "my dog ​​is apple" seen by the BERT model may be the replaced word. This forces the BERT model not to rely too much on the current word when encoding, but to consider the context of the word. For example, when encoding, the BERT model will encode the masked part as the semantics of "hairy" instead of "apple" based on the context of "my dog ​​is".

[0110] In the third part, the BERT model learns the association between sentences. Sentences usually have a certain association, so we hope that the BERT model can learn this association. The BERT model will extract two related sentences from Wikipedia with a probability of 50%. Here, related sentences refer to mutually related word sequences. The BERT model will also extract two unrelated sentences with a probability of 50%, and then the BERT model will determine whether the two sentences are related.

[0111] Part 4: Further training of the BERT model (Fine-Tuning).

[0112] For general classification tasks, see Figure 3 As shown in the figure, the input is a sequence TOK1~TOKN, all TOKEN belong to the same sentence, and then the last layer C of the special vocabulary [CLS] is used for smoothing to achieve classification, and the classified data is used to further train the BERT model.

[0113] For the task of inputting two sequences and calculating the similarity between the two sequences, the similarity task is to calculate the similarity between two words, see Figure 4 As shown in the figure, the words in the two sequences belong to different sentences, for example, TOK1~TOKN belong to sentence 1, and TOK1~TOKM belong to sentence 2. They are then classified using the last layer C of the special vocabulary [CLS], and the classified data is used to further train the BERT model.

[0114] For the task of sequence labeling, sequence labeling is to label the vocabulary with the corresponding label, see Figure 5 As shown in the figure, the input is a sequence TOK1~TOKN. Each word except [CLS] and [SEP] will have an output label at each moment. For example, the label of B-PER represents the word TOK2. The input label is then used to further train the BERT model.

[0115] Based on the similarity calculation task in the above task, the BERT model calculates the similarity between the query word and the keywords in the game introduction, and then uses the keyword with the maximum similarity with the query word as the game tag.

[0116] The BERT model can use the MASK model to make the vocabulary take into account the contextual meaning, and can determine whether two sentences are related by learning the association between sentences; through further training, it can calculate the similarity between query words and keywords, which means that the BERT model can solve the defects in the above solutions. However, since the BERT model has less training data and most of it is in English, the BERT model cannot calculate the similarity between Chinese query words and keywords well, and it cannot obtain accurate game tags in Chinese.

[0117] In the above schemes, TF-IDF and BM25 algorithms are based on traditional machine learning, and use keywords with high frequency as labels. However, TF-IDF and BM25 algorithms cannot identify polysemous words, resulting in low accuracy of the obtained labels. Word2Vec model and DSSM model convert structured words into word vectors, converting real-life problems into mathematical problems, opening up a new path for deep learning models. The semantic similarity between the word vector of the query word and the word vector of the keyword in the game introduction can be calculated, and the keyword with the greatest semantic similarity can be used as a label, so as to use the calculated label to verify the manually set label. Although it can solve the problem of polysemous words, it can only distinguish polysemous words under a single category, but cannot distinguish polysemous words under different categories, and the generalization ability of the model is low. Finally, the BERT model appeared, which can solve the problems of all the above algorithms, but the BERT model is suitable for English label prediction, and it cannot be applied to Chinese label prediction.

[0118] That is to say, adopting the above solution will result in the accuracy of the obtained game tags being low, and they cannot accurately describe the games. Then, when the user actually enters a query term in the app store, the accuracy of the games recalled based on the wrong game tags will also be low, and there will be certain errors in the recommended games.

[0119] In the related art, after a user enters a query word, although the application searched by the query word can be displayed in the first priority of the application store, the correlation between the related applications displayed in the subsequent priorities of the first priority and the application in the first priority is low, resulting in a low correlation between the application recommended to the user and the user's needs.

[0120] For example, see Figure 6As shown, if the above scheme is adopted, a first mapping relationship between the query word "Game A" and the game tag "Strategy" can be established. When the user enters the query word "Game A", it will be mapped to the game tag "Strategy", and then the game will be recalled according to the game tag "Strategy", so that the game "Game A" which is strongly related to the game tag "Strategy" can be recommended to the user in the first place, and the game "Game B" which has a relatively low correlation with the game tag "Strategy" can be recommended to the user in the second place. However, Game B and Game A are two different types of games. When the user enters the query word "Game A", he may want to find a team competition game similar to Game A, but the app store recommends Game B, a casual puzzle game, to the user in the second place. The correlation between the first-place game and the second-place game is low, and the correlation with the user's needs is also low.

[0121] Based on this, Figure 1 is a flowchart of a tag processing method according to an exemplary embodiment. Figure 1 As shown, the label processing method is used in a terminal and includes the following steps.

[0122] In step S11 , a first target tag for describing the application is determined.

[0123] The first target tags are tags in the original tag system of the application store. These tags can be obtained through a graphic recognition model. The first target tags include tags that can accurately describe the application.

[0124] In step S12, the text information of the application is used as input information of the chat program to obtain a second target tag for describing the application.

[0125] The text information of the application includes an application profile of the application, which can also be understood as a summary description of the application. The application profile is used to describe the name of the application, the features of the application, and the background of the application.

[0126] For example, if the application introduction is a game introduction, it will describe the name of the game, the company that produced the game, the game features, and the game background. The game features are reflected through game tags, such as whether the game is a casual, puzzle, action, or design game. The game background refers to the plot of the game and the era in which the plot takes place.

[0127] The chat program may be a chatgpt (Chat Generative Pre-trained Transformer) chat program, which may obtain an accurate second target tag describing the application program according to text information of the input application program.

[0128] For example, the labels related to "Game A" in the original application store are "multiplayer games" and "martial arts". In the present disclosure, the interface of the chat program can be called, and then "Please extract the keywords of the game from the following text: The world is vast and the world is not the only one! The mobile game "Game A" is a new national style, high-freedom martial arts action mobile game with the background of painted rivers and lakes. In Game A, a chaotic world with bandits in the late Tang Dynasty is created, and a high degree of freedom is open to martial arts exploration! In the game, you will immerse yourself in the free and unparalleled battles", and you can get "The world is vast and the world is not the only one", "Game A, mobile game, new national style, high degree of freedom, martial arts action, bandits in the late Tang Dynasty, chaotic rivers and lakes" and other second target labels that accurately describe Game A. It can be seen that the second target label output by the chat program can be more suitable for the application than the label of the original application.

[0129] The first target tag and the second target tag are keywords or phrases used to describe the content, features and type of the application in order to better classify, search and promote games. Different applications may have different tags. For example, if the application is a game and the tag is a game tag:

[0130] Game tags used to describe game types include: action, adventure, role-playing (RPG), shooting, racing, puzzle, simulation, strategy, etc.

[0131] Game tags used to describe the platforms on which a game can be run include: computer (PC), handheld, console (such as PlayStation, Xbox), mobile phone, etc.

[0132] Game tags used to describe the game's story background, plot or theme include: science fiction, fantasy, horror, history, war, etc.

[0133] Game tags used to describe the game's graphics style include: pixel style, cartoon style, realistic style, etc.

[0134] Game tags used to describe game mechanics include: open world, puzzle, survival, cards, etc.

[0135] Game tags used to describe the game mode include: single-player, cooperative, competitive, massively multiplayer online (MMO), etc.

[0136] Game tags used to describe the age groups for which the game is suitable include: E (all ages), T (teens), M (adults), etc.

[0137] Game tags used to describe the developer or publisher of a game include: the name of the game's development team or publishing company.

[0138] Game tags used to describe the language of the game include: the language versions supported by the game.

[0139] After obtaining the second target tags, the second target tags may be further screened to screen out valid second target tags.

[0140] For example, for the second target tags such as "Da Qian Jianghu, Game A, mobile game, new national style, high degree of freedom, martial arts action, bandits in the late Tang Dynasty, and Jianghu in troubled times", bandits in the late Tang Dynasty and Da Qian Jianghu can be filtered out, and the remaining Game A, mobile game, new national style, high degree of freedom, martial arts action, bandits in the late Tang Dynasty, and Jianghu in troubled times can be used as valid tags for Game A.

[0141] In step S13, the application and / or associated applications related to the application are recommended according to the first target tag and the second target tag corresponding to the query word input by the user.

[0142] After obtaining the first target tag and the second target tag of each application, a first mapping relationship between the query term and the first target tag and the second target tag can be established, and then based on the first mapping relationship, the first target tag and the second target tag corresponding to the query term currently input by the user can be determined, and the first target tag and the second target tag can be used to recall the application and / or associated applications.

[0143] The first mapping relationship is established in a manner including: collecting statistics of first target tags and second target tags corresponding to applications downloaded by historical query words, and establishing a first mapping relationship between the query words and the first target tags and the second target tags.

[0144] An application refers to an application that is highly correlated with the query term, for example, it may be the first-ranked application recommended by the app store; an associated application refers to an application that is less correlated with the query term than the application, for example, it may be an application recommended by the app store at a lower rank than the first-ranked application.

[0145] See also Figure 7 As shown, game A has first target tags and second target tags such as "role playing, cartoon style, single-player game", game B has first target tags and second target tags such as "role playing, cartoon style, multiplayer game"; game C has first target tags and second target tags such as "role playing, realistic style, competitive game".

[0146] When the user enters the query term "Magic", the first target tags and the second target tags that have the first mapping relationship with "Magic" are "Role-playing, Cartoon Style, Single-player Game". Since the number of different tags between the game tags of Game A and the game tags of the query term "Magic" is 0, Game A will be recommended to the user as the first priority. Since the number of different tags between the game tags of Game B and the corresponding game tags of the query term "Magic", which are "Role-playing, Cartoon Style, Single-player Game", is 1, Game B will be recommended to the user as the second priority. Since the number of different tags between the game tags of Game C and the corresponding game tags of the query term "Magic", which are "Role-playing, Cartoon Style, Single-player Game", is 2, Game C will be recommended to the user as the third priority.

[0147] It can be seen that when recommending application programs in the application store in different priorities, the first target tags and the second target tags corresponding to the query term will be determined first; then, the difference degrees between the tags of multiple application programs and the first target tags and the second target tags will be determined, and the application programs corresponding to the multiple tags will be arranged in descending order of the difference degrees to be recommended to the user.

[0148] Next, an example scenario will be used to explain the related technology and the present disclosure.

[0149] In the related technology, if the user enters the query term "Game A", and the tags corresponding to the query term "Game A" are "Strategy, Multi-player Game", and the games described by these tags are "Game A", "Game B", "Game C", then in addition to showing "Game A" in the first priority, "Game B", "Game C" and other games will also be shown in the subsequent priorities in turn. However, the recommended games such as "Game B" and "Game C" are not the same as the type of game "Game A" required by the user. For example, the type of Game A is a 5V5 MOBA team competitive game, while Game B is a 5V5 TPS shooting competitive game, and Game C is a casual puzzle game. The types of Game B and Game C are different from the type of Game A and also different from the type of game required by the user. At this time, the user experience will be reduced.

[0150] In this application, if the user enters the query term "Game A", and the tags corresponding to the query term "Game A" are "Strategy, Competitive, 5V5, MOBA", the obtained tags are more suitable for games of the type "Game A" and other types. Then, "Game A" will be shown in the first priority, and "Game D", "Game E" and other games strongly related to "Game A" will be shown in the subsequent priorities in turn. The game types of "Game D" and "Game E" are both 5V5 MOBA team competitive games. The recommended games can better meet the user's needs, thereby improving the user experience.

[0151] In addition to recommending the applications corresponding to the first target tag and the second target tag to the user, the first target tag and the second target tag are also displayed on the relevant page of the recommended application so that the user can understand the application through the first target tag and the second target tag.

[0152] Through the above technical solution, the text information of the application can be input into the chat program to obtain a second target tag that can more accurately describe the application.

[0153] First, as the number of tags corresponding to a single application increases, from the previous first target tag to the first target tag and the second target tag, the number of tags corresponding to the query terms increases, so the applications and related applications described by more tags will also be more precise. For example, the original tag corresponding to the query term "Game A" was "Multiplayer Game", and there are many applications corresponding to Multiplayer Game, and it is impossible to know which application among the multiple applications it is. Now the tags corresponding to the query term "Game A" include "Multiplayer Game, Competition, MOBA, Strategy", and the number of tags is more, so the number of applications described by more tags will be fewer, and naturally more precise.

[0154] Secondly, since the chat program can output a second target tag that is more closely related to the application, when the user enters a query term, the first target tag and the second target tag corresponding to the current query term will be obtained based on the first mapping relationship between the query term and the first target tag and the second target tag. The application described by the second target tag will also be more accurate and naturally more in line with user needs.

[0155] The specific embodiments and optional embodiments involved in the above-mentioned step S11 and step S12 are introduced below.

[0156] In the TF-IDF, BM25 model, Word2Vec model, DSSM model and BERT model, the text information of the application is used to obtain the application label. In addition to the text information, the application introduction also contains the image information of the application. If the application label is obtained only based on the text information, its accuracy is relatively low. Therefore, in addition to using the text information of the application, the present disclosure also uses the image information to obtain the application label, as follows:

[0157] The graphic and text information of the application is used as input information of a graphic and text recognition model to obtain the first target label, wherein the graphic and text information is used to describe the application in the form of pictures and texts.

[0158] See also Figure 8As shown, the graphic information of the application has two forms: pictures and texts. The graphic information includes: basic data of the application, search data and tag data. The basic data includes the application introduction and the application details. The application introduction is used to briefly introduce the application, for example Figure 8 The introduction in the lower left corner; the detailed image can be the first frame of the video in the application, for example Figure 8 Image A in the figure. Search data refers to the query words entered by users for the application. The query words used by users when searching for high-level applications can be counted and these query words are used as search data. Label data refers to the original label corresponding to the application, such as Figure 8 The original tags in the game include "competitive, auto chess", etc.

[0159] For example, see Fig. 9 As shown in the figure, the Contrastive Language-Image Pretraining (CLIP) model is to understand and generate text related to the image or images related to the text by learning a large amount of text and images. The text recognition model includes the following aspects:

[0160] First, the composition of the image and text recognition model. The image and text recognition model consists of two parts: the visual model and the language model. The visual model is usually a convolutional neural network, such as ResNet, which is used to extract features from image information; the language model is usually a conversion model, which is used to process and understand text information. The visual model and the language model share the weights of the network and are trained through contrastive learning.

[0161] Secondly, the image and text recognition model is trained through contrastive learning. The goal of contrastive learning is to enable the image and text recognition model to map similar images and texts to similar vector spaces, and to map dissimilar images and texts to distant vector spaces. When training the image and text recognition model, for each paired image sample and text sample, the image and text recognition model identifies whether the paired image sample and text sample match. Contrastive learning usually assigns higher scores to correctly paired image samples and text samples, and lower scores to incorrectly paired image samples and text samples.

[0162] The third aspect is the application of image and text recognition models. Image and text recognition models can understand the relationship between images and texts. For example, image and text recognition models can perform zero-sample learning, classifying images and texts of unknown categories without seeing any samples of any category. They can also generate images by recognizing the description of the text to generate the image corresponding to the text.

[0163] Therefore, the image and text recognition model has the ability to recognize image and text information. After taking the image and text information of the application as the input information of the image and text recognition model, the text and image can be encoded separately, and then the application vector and label vector corresponding to the application can be obtained. The application vector is the vector corresponding to the entire application, and the label vector is the vector corresponding to the first target label in the text information of the application.

[0164] Through the above technical solution, the image and text information of the application is used as the input information of the image and text recognition model to obtain the first target label, so that the obtained first target label takes into account the image information and text information of the application, thereby making the obtained first target label more accurate, and naturally the application described by the first target label will also be more accurate.

[0165] After obtaining the first target tag and the second target tag, it is necessary to establish a second mapping relationship between the first target tag and the application and a third mapping relationship between the second target tag and the application. In this way, after determining the first target tag and the second target tag corresponding to the query term, the corresponding application and / or associated application can be obtained according to the second mapping relationship and the third mapping relationship. Therefore, the following will specifically explain how to obtain the second mapping relationship and the third mapping relationship.

[0166] After obtaining the first target tag, it is also necessary to establish a second mapping relationship between the first target tag and the application, so that after determining the first target tag corresponding to the query term, the application corresponding to the first target tag and the associated application can be determined according to the second mapping relationship. Establishing the second mapping relationship includes: using the image and text information of the application as input information of the image and text recognition model to obtain the first tag and the application; clustering multiple applications and dividing them into multiple application groups; for each of the application groups, determining the first target tag that matches the application group from the multiple first tags.

[0167] The first label is the label corresponding to the label vector output by the image and text recognition model.

[0168] The application is the application corresponding to the application vector output by the image and text recognition model.

[0169] The first target tag is used to describe the application programs in the application program group, and a second mapping relationship exists between the first target tag and each application program in the application program group.

[0170] Multiple cluster center points can be randomly set first, and multiple application vectors can be clustered into multiple application groups centered on the multiple cluster center points; for each of the multiple application groups, the similarity between each application group and the first label is calculated, and the first label with the largest similarity is used as the first target label matching the application group, and the first target label has a second mapping relationship with each application in the application group.

[0171] For example, see Fig.10 As shown, there are currently five first labels "labels A to E" and 10 applications "applications" A to J. Five cluster centers can be randomly set first, and applications A to J can be clustered into five application groups, namely "applications A and B", "applications C and D", "applications E and F", "applications G and H", and "applications I and J"; then the similarities between the five first labels A to J and the five application groups "applications A and B", "applications C and D", "applications E and F", "applications G and H", and "applications I and J" can be calculated respectively.

[0172] Finally, it is determined that the similarity between label A and the application group "application A and B" is the greatest, the similarity between label B and the application group "application C and D" is the greatest, the similarity between label C and the application group "application E and F" is the greatest, the similarity between label D and the application group "application G and H" is the greatest, and the similarity between label E and the application group "application I and J" is the greatest. Thus, a second mapping relationship between different application groups and different first labels can be established.

[0173] In the above clustering scheme, the cluster center point can be determined according to the inflection point method, or the empirical value can be selected. As the cluster center point, n is the number of applications in the application cluster.

[0174] Through the above technical solution, a second mapping relationship between the first target tag and the application can be established. In this way, after obtaining the first target tag corresponding to the query word input by the user, the application and / or associated application corresponding to the first target tag can be obtained according to the second mapping relationship.

[0175] After obtaining the second target tag output by the chat program, it is also necessary to establish a third mapping relationship between the second target tag and the application, so that after determining the second target tag corresponding to the query word, the application corresponding to the second target tag and the associated application can be determined according to the third mapping relationship. Establishing the third mapping relationship includes: using the text information of the application as the input information of the chat program to obtain the second tag; clustering multiple applications and dividing them into multiple application groups; for each application group, determining the second target tag matching the application group from multiple second tags.

[0176] The second label is a label output by the chat program according to the text information of the application program.

[0177] The second target tag is used to describe an application in the application group, and a third mapping relationship exists between the second target tag and each application in the application group.

[0178] Multiple cluster center points can be randomly set first, and multiple application vectors can be clustered into multiple application groups centered on the multiple cluster center points; for each of the multiple application groups, the similarity between each application group and the second label is calculated, and the second label with the largest similarity is used as the second target label matching the application group, and the second target label has a third mapping relationship with each application in the application group.

[0179] Through the above technical solution, a third mapping relationship between the second target tag and the application can be established. In this way, after obtaining the second target tag corresponding to the query word input by the user, the application and / or associated application corresponding to the second target tag can be obtained according to the third mapping relationship.

[0180] In the process of determining the first target tag and the second target tag that match the application group, since the clustered application group may also contain applications that do not match the first target tag and the second target tag highly, it is necessary to further screen the applications in the application group, specifically including: from the multiple applications in the application group, screening out target applications whose distance from the center point of the application is less than a preset distance to form a target application group, and the first target tag is used to describe the applications in the target application group.

[0181] For example, a selection range can be defined with the cluster center of the application as the origin and a set distance as the radius to select the target application in the application group, and the target application in the selection range can form a new target application group. Through this technical solution, the applications in the selected target application group can have a high degree of matching with the first target tag and the second target tag.

[0182] After determining the first target tag that matches the application group, if the number of applications in the application group is small, it means that the first target tag can only describe a small number of applications and the quality of the first target tag is poor. In this case, the first target tag can be screened out. Specifically, the first target tag can be screened out when the number of applications mapped by the first target tag is less than a first preset number.

[0183] For example, if the first preset number is 2, the first target tag is a game tag, and the application is a game. If the number of games mapped to the game tag "Martial Arts Action" is only 1, it means that the quality of the game tag "Martial Arts Action" is poor. In this case, the "Martial Arts Action" tag can be removed and the game can be described without the "Martial Arts Action" tag.

[0184] By screening out the first target tag when the number of application programs mapped to the first target tag is less than a first preset number, some first target tags with poor quality can be removed.

[0185] After determining the first target tag that matches the application, if there are a large number of applications in the application, it means that the first target tag is relatively general and cannot reflect the characteristics of the application and cannot distinguish different applications. In this case, this part of the first target tag can be screened out. Specifically, the first target tag can be screened out when the number of applications mapped by the first target tag is greater than the second preset number.

[0186] For example, if the second preset number is 20, the first target tag is the game tag, and the application is a game. If the number of games mapped to the game tag "multiplayer games" is 50, it means that the game tag "multiplayer games" cannot distinguish different games individually. In this case, the "multiplayer games" tag can be removed, and the multiplayer game tag is not used to describe the game.

[0187] By filtering out the first target tag when the number of applications mapped to the first target tag is greater than the second preset number, some first target tags that cannot distinguish applications can be removed. And by filtering out the first target tag when the number of applications is less than the first preset number and the number of applications is greater than the second preset number, the number of applications mapped to the first target tag can be maintained within a reasonable range, that is, the number of applications mapped to the first target tag is neither too many nor too few, so that the applications recommended by the application store can also be maintained within a reasonable range.

[0188] Similarly, when the number of applications mapped to the second target tag is less than the first preset number or the number of applications mapped to the second target tag is greater than the second preset number, the second target tag can be screened out. The effect obtained is the same as that of the first target tag, which will not be repeated here.

[0189] Fig.11 FIG. 1 is a block diagram of a label processing device according to an exemplary embodiment. Fig.11 The label processing device 1100 includes: a first target label determination module 1120 , a second target label determination module 1130 and a recommendation module 1140 .

[0190] A first target tag determining module 1120 is configured to determine a first target tag for describing an application;

[0191] The second target tag determination module 1130 is configured to use the text information of the application as input information of the chat program to obtain a second target tag for describing the application;

[0192] The recommendation module 1140 is configured to recommend the application and / or associated applications related to the application according to the first target tag and the second target tag.

[0193] Optionally, the first target tag determining module 1120 includes:

[0194] The first prediction submodule is configured to use the image and text information of the application as input information of the image and text recognition model to obtain the first target label, wherein the image and text information is used to describe the application in the form of pictures and texts.

[0195] Optionally, the first prediction submodule includes:

[0196] A second prediction submodule is configured to use the image and text information of the application as input information of the image and text recognition model to obtain a first label and an application;

[0197] A clustering submodule is configured to cluster the plurality of applications into a plurality of application groups;

[0198] The first target tag determination submodule is configured to determine, for each application group, a first target tag matching the application group from a plurality of the first tags, wherein the first target tag is used to describe the application in the application group.

[0199] Optionally, the label processing device 1100 further includes:

[0200] The screening module is configured to screen out the first target tag when the number of applications is less than a first preset number or the number of applications is greater than a second preset number; the number of applications is the number of applications in the application group matching the first target tag.

[0201] Optionally, the label processing device 1100 further includes:

[0202] The screening module is configured to screen out target applications whose distance from the center point of the application is less than a preset distance from the multiple applications in the application group to form a target application group, and the first target tag is used to describe the applications in the target application group.

[0203] Optionally, the second target tag determining module 1130 includes:

[0204] A second label determination submodule is configured to use the text information of the application as input information of the chat program to obtain a second label;

[0205] The matching submodule is configured to determine, for each application group, a second target tag that matches the application group from a plurality of the second tags, where the second target tag is used to describe the application in the application group.

[0206] Optionally, the label processing device 1100 further includes:

[0207] The display module is configured to display the first target tag and the second target tag.

[0208] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0209] The present disclosure also provides a computer-readable storage medium having computer program instructions stored thereon, and the program instructions, when executed by a processor, implement the steps of the label processing method provided by the present disclosure.

[0210] Fig.12 1 is a block diagram of an apparatus 1200 for tag processing according to an exemplary embodiment. For example, the apparatus 1200 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0211] Reference Fig.12 , the device 1200 may include one or more of the following components: a first processing component 1202 , a first memory 1204 , a first power component 1206 , a multimedia component 1208 , an audio component 1210 , a first input / output interface 1212 , a sensor component 1214 , and a communication component 1216 .

[0212] The first processing component 1202 generally controls the overall operation of the device 1200, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The first processing component 1202 may include one or more processors 1220 to execute instructions to complete all or part of the steps of the above-mentioned tag processing method. In addition, the first processing component 1202 may include one or more modules to facilitate the interaction between the first processing component 1202 and other components. For example, the first processing component 1202 may include a multimedia module to facilitate the interaction between the multimedia component 1208 and the first processing component 1202.

[0213] The first memory 1204 is configured to store various types of data to support operations on the device 1200. Examples of such data include instructions for any application or method operating on the device 1200, contact data, phone book data, messages, pictures, videos, etc. The first memory 1204 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access first memory (SRAM), electrically erasable programmable read-only first memory (EEPROM), erasable programmable read-only first memory (EPROM), programmable read-only first memory (PROM), read-only first memory (ROM), magnetic first memory, flash first memory, magnetic disk or optical disk.

[0214] The first power source component 1206 provides power to various components of the device 1200. The first power source component 1206 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device 1200.

[0215] The multimedia component 1208 includes a screen that provides an output interface between the device 1200 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 1208 includes a front camera and / or a rear camera. When the device 1200 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0216] The audio component 1210 is configured to output and / or input audio signals. For example, the audio component 1210 includes a microphone (MIC), and when the device 1200 is in an operation mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the first memory 1204 or sent via the communication component 1216. In some embodiments, the audio component 1210 also includes a speaker for outputting audio signals.

[0217] The first input / output interface 1212 provides an interface between the first processing component 1202 and a peripheral interface module, which may be a keyboard, a click wheel, buttons, etc. These buttons may include but are not limited to: a home button, a volume button, a start button, and a lock button.

[0218] The sensor assembly 1214 includes one or more sensors for providing various aspects of the status assessment of the device 1200. For example, the sensor assembly 1214 can detect the open / closed state of the device 1200, the relative positioning of components, such as the display and keypad of the device 1200, the sensor assembly 1214 can also detect the position change of the device 1200 or a component of the device 1200, the presence or absence of user contact with the device 1200, the orientation or acceleration / deceleration of the device 1200, and the temperature change of the device 1200. The sensor assembly 1214 can include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 1214 can also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 1214 can also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0219] The communication component 1216 is configured to facilitate wired or wireless communication between the device 1200 and other devices. The device 1200 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 1216 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 1216 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0220] In an exemplary embodiment, the device 1200 can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above-mentioned label processing method.

[0221] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a first memory 1204 including instructions, and the instructions can be executed by a processor 1220 of the device 1200 to complete the above-mentioned tag processing method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access first memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0222] In addition to being an independent electronic device, the above-mentioned device can also be a part of an independent electronic device. For example, in one embodiment, the device can be an integrated circuit (IC) or a chip, wherein the integrated circuit can be an IC or a collection of multiple ICs; the chip can include but is not limited to the following types: GPU (Graphics Processing Unit), CPU (Central Processing Unit), FPGA (Field Programmable Gate Array), DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), SOC (System on Chip, SoC), etc. The above-mentioned integrated circuit or chip can be used to execute executable instructions (or codes) to implement the above-mentioned label processing method. The executable instructions can be stored in the integrated circuit or chip, or can be obtained from other devices or equipment, such as the integrated circuit or chip including a processor, a memory, and an interface for communicating with other devices. The executable instruction can be stored in the memory, and when the executable instruction is executed by the processor, the above-mentioned label processing method is implemented; alternatively, the integrated circuit or chip can receive the executable instruction through the interface and transmit it to the processor for execution, so as to implement the above-mentioned label processing method.

[0223] In another exemplary embodiment, a computer program product is also provided. The computer program product includes a computer program that can be executed by a programmable device. The computer program has a code portion for executing the above-mentioned tag processing method when executed by the programmable device.

[0224] Fig.13 1 is a block diagram of a device 1300 for tag processing according to an exemplary embodiment. For example, the device 1300 may be provided as a server. Fig.13 The apparatus 1300 includes a second processing component 1322, which further includes one or more processors, and a second memory resource represented by a second memory 1332 for storing instructions executable by the second processing component 1322, such as an application. The application stored in the second memory 1332 may include one or more modules each corresponding to a set of instructions. In addition, the second processing component 1322 is configured to execute instructions to perform the above-mentioned tag processing method.

[0225] The device 1300 may further include a second power supply component 1326 configured to perform power management of the device 1300, a wired or wireless network interface 1350 configured to connect the device 1300 to a network, and a second input / output interface 1358. The device 1300 may operate based on an operating system stored in the second memory 1332.

[0226] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the present disclosure. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0227] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A label processing method, characterized in that: include: determining a first target tag for describing the application; Using the text information of the application as input information of the chat program to obtain a second target tag for describing the application; The application and / or associated applications related to the application are recommended according to the first target tag and the second target tag corresponding to the query word input by the user.

2. The method according to claim 1, characterized in that The determining of a first target tag for describing the application comprises: The graphic and text information of the application is used as input information of a graphic and text recognition model to obtain the first target label, wherein the graphic and text information is used to describe the application in the form of pictures and texts.

3. The method according to claim 2, characterized in that The step of using the image and text information of the application as input information of the image and text recognition model to obtain the first target label includes: Using the image and text information of the application as input information of the image and text recognition model to obtain a first label and the application; Cluster multiple applications and divide them into multiple application groups; For each of the application groups, a first target tag matching the application group is determined from a plurality of the first tags, where the first target tag is used to describe the application in the application group.

4. The method according to claim 3, characterized in that The method further comprises: When the number of applications is less than a first preset number or the number of applications is greater than a second preset number, the first target tag is screened out; the number of applications is the number of applications in the application group that matches the first target tag.

5. The method according to claim 3, characterized in that: The method further comprises: Target applications whose distance from a center point of the application is less than a preset distance are screened out from the multiple applications in the application group to form a target application group, wherein the first target tag is used to describe the applications in the target application group.

6. The method according to claim 1, characterized in that The step of using the text information of the application as input information of the chat program to obtain a second target tag for describing the application includes: Using the text information of the application as input information of the chat program to obtain a second label; For each application group, a second target tag matching the application group is determined from a plurality of second tags, where the second target tag is used to describe the applications in the application group.

7. The method according to claim 1, characterized in that The method further comprises: The first target label and the second target label are displayed.

8. A label processing device, characterized in that: include: A first target tag determining module, configured to determine a first target tag for describing an application; A second target tag determination module is configured to use the text information of the application as input information of the chat program to obtain a second target tag for describing the application; The recommendation module is configured to recommend the application and / or associated applications related to the application according to the first target tag and the second target tag.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to: Execute the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the steps of the method described in any one of claims 1 to 7 are implemented.

Images