Information distribution method and device, electronic equipment and storage medium

Abstract

The invention provides an information distribution method and device, electronic equipment and a storage medium, and relates to the technical fields of Internet of Things, artificial intelligence, big data and the like. The method comprises the following steps: in response to an information query statement, performing first feature extraction on the information query statement to obtain a first information feature vector; scheduling at least one generative model instance from a generative model pool, and rewriting the first information feature vector to obtain at least one information rewriting statement corresponding to the information query statement; scheduling at least one discrimination model instance from a discrimination model pool, performing correlation analysis on the information query statement and the information rewriting statement to obtain a target rewriting statement of which the correlation meets a preset condition, the generation model pool and the discrimination model pool being deployed in the same physical machine; and performing information distribution according to the target rewriting statement.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, in particular to the fields of the Internet of Things, artificial intelligence, big data, and more specifically to an information distribution method, device, electronic device, and storage medium. Background Art

[0002] When a user needs to search for the information they want, they will enter their query information in the input box, and then click query, and the background will return the relevant information query results to the user.

[0003] In the process of realizing the concept of the present disclosure, the inventors discovered that there are at least the following problems in the related technology: due to the diversity of user expressions, there are differences between the user's query information and the benchmark description information. When the query information searched by the user does not match the benchmark description information, some information cannot be recalled and cannot be searched by the user. Summary of the Invention

[0004] In view of this, the present disclosure provides an information distribution method, apparatus, electronic device, and storage medium.

[0005] One aspect of the present disclosure provides an information distribution method, including: in response to an information query statement, performing a first feature extraction on the information query statement to obtain a first information feature vector; scheduling at least one generation model instance from a generation model pool, rewriting the first information feature vector to obtain at least one information rewriting statement corresponding to the information query statement; scheduling at least one discriminant model instance from a discriminant model pool, performing a correlation analysis on the information query statement and the information rewriting statement, and obtaining a target rewriting statement whose correlation meets a preset condition, wherein the generation model pool and the discriminant model pool are deployed in the same physical machine; and distributing information according to the target rewriting statement.

[0006] According to an embodiment of the present disclosure, performing a first feature extraction on an information query statement to obtain a first information feature vector includes: performing a first tokenization process on the information query statement to obtain a first string sequence determined for the information query statement; performing a first vectorization process on the first string sequence to obtain a first string identification feature sequence and a first string position feature sequence of the first string sequence; and using a first summation operator to perform a sum calculation on the first string identification feature sequence and the first string position feature sequence to obtain the first information feature vector.

[0007] According to an embodiment of the present disclosure, rewriting the first information feature vector to obtain at least one information rewriting statement corresponding to the information query statement includes: performing a first encoding process on the first information feature vector to obtain at least one rewritten string identification sequence corresponding to the original string sequence of the information query statement; querying an existing vocabulary table based on the at least one rewritten string identification sequence to obtain at least one rewritten string sequence, wherein a one-to-one mapping relationship between the rewritten string identification and the rewritten string is predefined in the existing vocabulary table; and generating at least one information rewriting statement based on the at least one rewritten string sequence.

[0008] According to an embodiment of the present disclosure, generating at least one information rewriting statement based on at least one rewritten string sequence includes: deduplicating the rewritten string sequence to obtain a first string sequence; comparing the first string sequence with a blacklist string, deleting a target string in the first string sequence that matches the blacklist string to obtain a second string sequence; and combining the second string sequences to generate the information rewriting statement.

[0009] According to an embodiment of the present disclosure, a correlation analysis is performed on an information query statement and an information rewriting statement to obtain a target rewriting statement whose correlation meets a preset condition, including: performing a second feature extraction on a combination statement of the information query statement and any one of the information rewriting statements to obtain a second information feature vector; performing a second encoding process on the second information feature vector to obtain a target correlation value between the information query statement and the information rewriting statement; and determining the information rewriting statement whose target correlation value meets the preset condition as the target rewriting statement.

[0010] According to an embodiment of the present disclosure, performing a second feature extraction on a combination statement of an information query statement and any one of the information rewriting statements to obtain a second information feature vector includes: performing a second tokenization process on the combination statement of an information query statement and any one of the information rewriting statements to obtain a second string sequence determined for the combination statement; performing a second vectorization process on the second string sequence to obtain a second string identification feature sequence and a second string position feature sequence of the second string sequence; and using a second summation operator to perform a sum calculation on the second string identification feature sequence and the second string position feature sequence to obtain a second information feature vector.

[0011] According to an embodiment of the present disclosure, performing a second encoding process on the second information feature vector to obtain a target correlation value between the information query statement and the information rewriting statement includes: performing a second encoding process on the second information feature vector to generate multiple rewriting types of the information rewriting statement relative to the information query statement and candidate correlation values ​​corresponding to each rewriting type to obtain multiple candidate correlation values; and determining the candidate correlation value with the largest value among the multiple candidate correlation values ​​as the target correlation value.

[0012] Another aspect of the present disclosure provides an information distribution device, including: a first feature extraction module, used to perform a first feature extraction on the information query statement in response to the information query statement, and obtain a first information feature vector; a rewriting module, used to schedule at least one generation model instance from the generation model pool, rewrite the first information feature vector, and obtain at least one information rewriting statement corresponding to the information query statement; a correlation analysis module, used to schedule at least one discriminant model instance from the discriminant model pool, perform correlation analysis on the information query statement and the information rewriting statement, and obtain a target rewriting statement whose correlation meets a preset condition, wherein the generation model pool and the discriminant model pool are deployed in the same physical machine; and an information distribution module, used to distribute information according to the target rewriting statement.

[0013] Another aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the information distribution method of the present disclosure.

[0014] Another aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the information distribution method of the present disclosure.

[0015] Another aspect of the present disclosure provides a computer program product, which includes computer-executable instructions. When the instructions are executed, the computer program product is used to implement the information distribution method of the present disclosure.

[0016] According to the embodiments of the present disclosure, a first feature extraction is performed on the information query statement in response to the information query statement to obtain a first information feature vector; at least one generation model instance is scheduled from the generation model pool to rewrite the first information feature vector to obtain at least one information rewrite statement corresponding to the information query statement; at least one discriminant model instance is scheduled from the discriminant model pool to perform correlation analysis on the information query statement and the information rewrite statement to obtain a target rewrite statement whose correlation meets the preset conditions, wherein the generation model pool and the discriminant model pool are deployed in the same physical machine; and a technical means of distributing information according to the target rewrite statement is adopted. Since the generation model and the discriminant model are deployed on the same machine within a single service, the efficient reasoning of the two complex models is realized through the model pooling deployment and model orchestration scheduling capabilities, and the online real-time rewriting of the information query statement is completed. When there is a difference between the user's query information and the benchmark description information, the rewritten statement can be combined to search for user demand information that matches the user's needs and is more complete and rich, which can effectively improve the completeness, accuracy and user experience of the search results. Furthermore, by combining feature extraction with model processing, we can reduce consideration of query length and focus on semantic information. This makes it easier to generate rewritten queries for long-tail queries, effectively addressing the difficulty of generating rewritten long-tail queries online in real time. Pooling the generative and discriminative models on the same machine fully utilizes GPU computing power, improves inference service throughput, and eliminates the time-consuming RPC communication associated with deploying the two models independently. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0018] Figure 1 Schematically illustrates an exemplary system architecture to which the information distribution method according to an embodiment of the present disclosure can be applied;

[0019] Figure 2 The following schematically shows a flow chart of an information distribution method according to an embodiment of the present disclosure;

[0020] Figure 3 The following schematically illustrates a schematic diagram of online deployment of a model according to an embodiment of the present disclosure;

[0021] Figure 4A A schematic diagram schematically illustrates a process of calling a BART instance to obtain a rewritten string identification sequence according to an embodiment of the present disclosure;

[0022] Figure 4BA schematic diagram schematically illustrates a process of invoking a BERT instance to obtain a relevance score of a rewritten query relative to a query according to an embodiment of the present disclosure;

[0023] Figure 5 Schematically shows the overall flow chart of online real-time generation of rewritten query according to an embodiment of the present disclosure;

[0024] Figure 6 A block diagram schematically illustrates an information distribution device according to an embodiment of the present disclosure; and

[0025] Figure 7 A block diagram schematically shows an electronic device suitable for implementing the information distribution method according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0026] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0027] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0029] When expressions such as "at least one of A, B, and C, etc." are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0030] In the embodiments of this disclosure, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of all data involved (including, but not limited to, user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information and maintain information security.

[0031] In the embodiments of the present disclosure, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0032] Related information matching technologies primarily include precise, phrase, word segmentation, and intelligent matching. Rewriting user search queries primarily focuses on word segmentation matching. Synonymically rewriting queries can improve word matching, thereby increasing CPC (cost per click). Rewriting queries to synonyms can alleviate insufficient information recall, enrich search results, increase RPM (revenue per thousand), and improve the user experience.

[0033] In related art, query rewriting is primarily achieved by online loading of an offline-generated query rewriting vocabulary that stores mappings between original queries and rewritten queries. This can cover approximately 80% of online user search queries. During the implementation of the present disclosure, the inventors discovered that, due to vocabulary size limitations and traffic fluctuations, the query rewriting function cannot cover all queries, making it difficult to generate rewritten queries for some long-tail queries.

[0034] Embodiments of the present disclosure provide an information distribution method, apparatus, electronic device, and storage medium, the method comprising: in response to an information query statement, performing a first feature extraction on the information query statement to obtain a first information feature vector; scheduling at least one generation model instance from a generation model pool, rewriting the first information feature vector, and obtaining at least one information rewriting statement corresponding to the information query statement; scheduling at least one discriminant model instance from a discriminant model pool, performing a correlation analysis on the information query statement and the information rewriting statement, and obtaining a target rewriting statement whose correlation meets a preset condition, wherein the generation model pool and the discriminant model pool are deployed in the same physical machine; and distributing information according to the target rewriting statement.

[0035] Figure 1 The following schematically illustrates an exemplary system architecture 100 to which the information distribution method according to an embodiment of the present disclosure can be applied. Figure 1The examples shown are merely examples of system architectures to which the embodiments of the present disclosure may be applied, to help those skilled in the art understand the technical content of the present disclosure, but do not mean that the embodiments of the present disclosure may not be used in other devices, systems, environments or scenarios.

[0036] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0037] A user may use a first terminal device 101, a second terminal device 102, or a third terminal device 103 to interact with a server 105 via a network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, or the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software (for example only).

[0038] The first terminal device 101 , the second terminal device 102 , and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0039] The server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process received data such as user requests, and feed back processing results (e.g., web pages, information, or data obtained or generated based on user requests) to the terminal devices.

[0040] It should be noted that the information distribution method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the information distribution device provided in the embodiment of the present disclosure can generally be set in the server 105. The information distribution method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the information distribution device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Alternatively, the information distribution method provided in the embodiment of the present disclosure can also be executed by the first terminal device 101, the second terminal device 102 or the third terminal device 103, or can also be executed by other terminal devices different from the first terminal device 101, the second terminal device 102 or the third terminal device 103. Accordingly, the information distribution device provided in the embodiment of the present disclosure can also be set in the first terminal device 101, the second terminal device 102 or the third terminal device 103, or in other terminal devices different from the first terminal device 101, the second terminal device 102 or the third terminal device 103.

[0041] For example, the information query statement may be originally stored in any one of the first terminal device 101, the second terminal device 102, or the third terminal device 103 (for example, the first terminal device 101, but not limited thereto), or stored on an external storage device and imported into the first terminal device 101. The first terminal device 101 may then locally execute the information distribution method provided by the embodiment of the present disclosure, or send the information query statement to other terminal devices, servers, or server clusters, and the other terminal devices, servers, or server clusters that receive the information query statement may execute the information distribution method provided by the embodiment of the present disclosure.

[0042] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0043] Figure 2 The flowchart of the information distribution method according to the embodiment of the present disclosure is schematically shown.

[0044] like Figure 2 As shown, the method includes operations S201 to S204.

[0045] In operation S201 , in response to an information query statement, a first feature extraction is performed on the information query statement to obtain a first information feature vector.

[0046] According to an embodiment of the present disclosure, an information query statement may represent a query input by a user when searching for information. A first information feature vector may represent a vectorized representation of the query. The first feature extraction process may be implemented using any model or module with feature extraction capabilities, without limitation herein.

[0047] In operation S202, at least one generation model instance is dispatched from the generation model pool, and a rewriting process is performed on the first information feature vector to obtain at least one information rewriting sentence corresponding to the information query sentence.

[0048] According to an embodiment of the present disclosure, a plurality of generation model instances may be pre-configured in the generation model pool, and each generation model instance may independently implement the rewriting processing operation. The rewriting process may be implemented using a pre-trained model with a corresponding rewriting function; or a large language model may be used, and implemented by adding prompt words related to "rewriting" on the basis of the information query statement, which is not limited here. The information rewriting statement may represent a rewritten query obtained by performing at least one of the following rewriting processes on the query query or part of the query words therein, such as synonym rewriting, upper rewriting, lower rewriting, etc.

[0049] For example, a query generation model can be pre-trained based on an algorithm and used as a generation model instance to generate one or more rewritten queries based on a query such as a title or attribute. The rewriting process can be implemented using the query generation model. For example, the query generation model can use Bidirectional and Auto-Regressive Transformers (BART), but is not limited to this model.

[0050] In operation S203, at least one discriminant model instance is scheduled from the discriminant model pool, and a correlation analysis is performed on the information query statement and the information rewriting statement to obtain a target rewriting statement whose correlation meets a preset condition, wherein the generation model pool and the discriminant model pool are deployed in the same physical machine.

[0051] According to an embodiment of the present disclosure, a plurality of discriminant model instances may be pre-configured in the discriminant model pool, and each discriminant model instance may independently implement the correlation analysis operation. Correlation analysis may be used to analyze and obtain the correlation score between the information query statement and the information rewriting statement. The correlation analysis process may be implemented using a pre-trained model with corresponding correlation analysis functions; or a large language model may be used, and implemented by adding prompt words related to "outputting the degree of correlation between the two" on the basis of the information query statement and the information rewriting statement, which is not limited here. The preset condition may represent that the correlation score meets a preset range, and may not be limited thereto.

[0052] For example, a query discriminant model can be pre-trained based on the algorithm and used as a discriminant model instance to calculate the relevance score between each rewritten query and the original query. Correlation analysis can be implemented using the query discriminant model. For example, the query discriminant model can use BERT (Bidirectional Encoder Representation from Transformers), but is not limited to this model.

[0053] According to the embodiments of the present disclosure, due to online time constraints, the generative model and the discriminative model need to be deployed on the same machine. Deploying the generative model pool and the discriminative model pool on the same physical machine means that the generative model instances and the discriminative model instances can use the same physical machine resources and directly exchange information without the need for remote communication such as RPC (Remote Procedure Call Protocol).

[0054] Figure 3 The figure schematically shows a schematic diagram of online deployment of a model according to an embodiment of the present disclosure.

[0055] like Figure 3 As shown, the generative model and the discriminative model can be deployed in the inference engine 300 in a pooled manner on the same machine. During online deployment, multiple instances of BART and BERT can be deployed separately through configurable parameters to form a generative model pool 310 and a discriminative model pool 320.

[0056] According to the embodiments of the present disclosure, based on the inference dependencies of the business logic control model, a configuration-based model orchestration capability can be implemented. For example, when an online service receives a request, the inference engine 300 can dispatch an instance from the generation model pool 310 to complete the generation of multiple rewritten queries, and then dispatch an instance from the discrimination model pool 320 to complete the calculation of the relevance score between the rewritten query and the query. Multiple model instances in the generation model pool 310 and the discrimination model pool 320 can be executed concurrently to process the online request, achieving a multi-stream effect and improving inference performance.

[0057] It should be noted that the above-mentioned generative model instance and discriminative model instance can also be replaced by other models or other integrated algorithm programs with corresponding functions, which are not limited here.

[0058] In operation S204, the statement is rewritten according to the target and information is distributed.

[0059] According to embodiments of the present disclosure, when distributing information for an information query, in addition to retrieving information related to the information query, information related to the target rewritten statement may also be retrieved and distributed simultaneously. The information ultimately distributed may include not only information related to the information query, but also information related to the target rewritten statement.

[0060] Through the above-mentioned embodiments of the present disclosure, the generative model and the discriminant model are deployed on the same machine within a single service. Through the ability of model pooling deployment and model orchestration scheduling, efficient reasoning of the two complex models is achieved, and the online real-time rewriting of information query statements is completed. When there is a difference between the user's query information and the benchmark description information, the rewritten statement can be combined to search for more complete and rich user demand information that matches the user's needs, which can effectively improve the completeness, accuracy and user experience of the search results. In addition, by combining feature extraction with model processing, the consideration of the sentence length of the information query statement can be reduced, and the focus can be placed on the consideration of the sentence semantic information, so that it is also easy to generate rewritten queries for long-tail queries, which can effectively solve the problem that long-tail queries are difficult to generate and rewrite online in real time. The pooled deployment of the generative model and the discriminant model on the same machine can make full use of the computing power of the GPU, improve the throughput of the inference service, and eliminate the RPC communication time consumed by the independent deployment of the two models.

[0061] In conjunction with specific embodiments, Figure 2 The method shown is further explained.

[0062] According to an embodiment of the present disclosure, operation S201 may include: performing a first tokenization process on the information query statement to obtain a first string sequence determined for the information query statement; performing a first vectorization process on the first string sequence to obtain a first string identification feature sequence and a first string position feature sequence of the first string sequence; and using a first summation operator to sum the first string identification feature sequence and the first string position feature sequence to obtain a first information feature vector.

[0063] According to an embodiment of the present disclosure, the first tokenization process is similar to the word segmentation process, and the information query statement can be divided into multiple units by means of tags. Each unit can be a first character string determined by a word, text or other symbol, which is not limited here. The first vectorization process can realize the conversion of each first character string into a vector with a fixed dimension. The vector may include a first character string identification feature representing the semantic information of the first character string and a first character string position feature representing the position information of the first character string in the information query statement. The first summation operator can represent a pre-edited summation program, which is used to combine the embedding process to perform sum calculation on the first character string identification feature sequence and the first character string position feature sequence to obtain a first information feature vector.

[0064] For example, after a user's search query is submitted to the online model service, the query can first be tokenized and corresponding feature information can be calculated, including the token ID and position of each first string in the query. This feature information can then be summed to obtain the first information feature vector corresponding to the query.

[0065] Through the above-described embodiments of the present disclosure, by obtaining the first string identification feature sequence and the first string position feature sequence, it is possible to extract relatively complete features of the information query statement. Using the first summation operator, a high degree of operator fusion and customized processing of some steps in the reasoning process are achieved.

[0066] According to an embodiment of the present disclosure, operation S202 may include: performing a first encoding process on the first information feature vector to obtain at least one rewritten string identifier sequence corresponding to the original string sequence of the information query statement; querying an existing vocabulary table based on the at least one rewritten string identifier sequence to obtain at least one rewritten string sequence, wherein the existing vocabulary table predefines a one-to-one mapping relationship between rewritten string identifiers and rewritten strings; and generating at least one information rewriting statement based on the at least one rewritten string sequence.

[0067] According to an embodiment of the present disclosure, the first encoding process may be implemented by calling a BART instance. By inputting the first information feature vector of the information query statement into the BART instance, at least one rewritten character string identification sequence may be output.

[0068] Figure 4A A schematic diagram schematically illustrates the process of calling a BART instance to obtain a rewritten string identification sequence according to an embodiment of the present disclosure.

[0069] like Figure 4AAs shown, before BART instance processing, the query token IDs (the token ID sequence of the first string in the query), position, etc. can be used as input to the first embedding 410 in combination with the first embedding 410, outputting a first information feature vector. During BART instance processing, the first information feature vector can be used as input to the first encoder 411. Combined with the self-attention calculation and feedforward network processing in the first encoder 411, and the feedforward network processing, cross-attention calculation, and self-attention calculation in the decoder 412, the decoder 412 outputs multiple token IDs of the rewritten query (the token ID sequence of the rewritten string in the rewritten query).

[0070] According to an embodiment of the present disclosure, in combination with an existing vocabulary, a rewritten string sequence is determined according to each rewritten string identification sequence, and the strings in the rewritten string sequence can be combined and concatenated to obtain an information rewriting sentence corresponding to the rewritten string sequence.

[0071] For example, a BART instance can perform multiple encoder-decoder calculations on the input first information feature vector and output the results. The model output is a sequence of token IDs for rewriting the string in multiple rewritten queries. The multiple token ID sequences output by the BART instance are combined into at least one query by querying the segmentation vocabulary, thereby obtaining at least one rewritten query.

[0072] Through the above-mentioned embodiments of the present disclosure, a rewritten string identifier sequence is determined by combining the first information feature vector of the information query statement, and then a method for determining the rewritten string sequence for constructing the information rewriting statement is determined. Since it is only necessary to maintain a vocabulary table about identifiers-strings in the string dimension, and there is no need for a vocabulary table about query statements-rewritten statements in the sentence dimension, it can effectively alleviate the problem that the vocabulary space is large and cannot cover all query statements-rewritten statements, and can be beneficial for rewriting long-tail queries.

[0073] In the fundamentally disclosed embodiment, generating at least one information rewriting statement based on at least one rewritten string sequence may include: removing duplicates from the rewritten string sequence to obtain a first string sequence; comparing the first string sequence with blacklisted strings, deleting target strings from the first string sequence that match the blacklisted strings to obtain a second string sequence; and combining the second string sequences to generate the information rewriting statement.

[0074] According to an embodiment of the present disclosure, after combining the second string sequence, the combination results may be first deduplicated and compared with the strings in the blacklist. After removing the combination results in the blacklist, the remaining combination results may be determined as information rewriting statements.

[0075] For example, after obtaining the rewritten queries, the rewritten queries may be subjected to deduplication and blackword filtering as a data set for subsequent processing.

[0076] Through the above-mentioned embodiments of the present disclosure, low-quality and repetitive sentences can be filtered out, which is conducive to improving the quality of rewritten sentences.

[0077] According to an embodiment of the present disclosure, operation S203 may include: performing a second feature extraction on a combination of the information query statement and any one of the information rewriting statements to obtain a second information feature vector; performing a second encoding process on the second information feature vector to obtain a target correlation value between the information query statement and the information rewriting statement; and determining an information rewriting statement whose target correlation value satisfies a preset condition as a target rewriting statement.

[0078] It should be noted that the second feature extraction can have the same or similar implementation method as the first feature extraction, which will not be described in detail here.

[0079] According to an embodiment of the present disclosure, performing a second feature extraction on the combination of the information query statement and any one of the information rewriting statements to obtain a second information feature vector may include: performing a second tokenization process on the combination of the information query statement and any one of the information rewriting statements to obtain a second string sequence determined for the combination; performing a second vectorization process on the second string sequence to obtain a second string identification feature sequence and a second string position feature sequence of the second string sequence; and using a second summation operator to sum the second string identification feature sequence and the second string position feature sequence to obtain the second information feature vector.

[0080] It should be noted that the second tokenization process can be implemented in the same or similar manner as the first tokenization process, the second vectorization process can be implemented in the same or similar manner as the first vectorization process, and the second summation operator can be implemented in the same or similar manner as the first summation operator, which will not be further described here. However, since the second tokenization process processes a composite sentence, after the second vectorization process, parameters representing the composite feature can also be obtained based on the second string identification feature sequence and the second string position feature sequence.

[0081] For example, the query and the generated set of rewritten queries can first be tokenized and corresponding feature information calculated, including the token ID, segment (representing the combined feature), and position of the second string in the combined query of the query and each rewritten query. This feature information can then be summed to obtain a second information feature vector corresponding to the combined query.

[0082] According to an embodiment of the present disclosure, the second encoding processing process can be implemented by calling a BERT instance. By inputting the second information feature vector of the information query sentence into the BERT instance, the target correlation value between each information rewriting sentence and the information query sentence can be output.

[0083] For example, the second information feature vector may be used as input to a BERT instance, and the BERT instance may calculate the relevance score between the query and each rewritten query based on the input second information feature vector.

[0084] According to an embodiment of the present disclosure, for example, a rewritten query having a relevance score with respect to the query that is greater than or equal to a preset threshold may be determined as a target rewritten statement, and the condition is not limited to this.

[0085] Through the above-described embodiments of the present disclosure, by performing a correlation analysis on the features of the combined information query statement and any information rewriting statement, it is advantageous to more accurately obtain the correlation value between the two, thereby improving the accuracy of the target rewriting statement obtained. The use of the second summation operator achieves a high degree of operator integration and customized processing of some steps in the reasoning process.

[0086] According to an embodiment of the present disclosure, performing the second encoding process on the second information feature vector to obtain the correlation value between the information query statement and the information rewriting statement may include: performing the second encoding process on the second information feature vector to generate multiple rewriting types of the information rewriting statement relative to the information query statement and candidate correlation values ​​corresponding to each rewriting type, thereby obtaining multiple candidate correlation values. Determining the candidate correlation value with the largest value among the multiple candidate correlation values ​​as the target correlation value.

[0087] According to an embodiment of the present disclosure, the rewriting type may include at least one of the following: synonymous rewriting, superordinate rewriting, subordinate rewriting, rewriting error, etc., and may not be limited thereto.

[0088] Figure 4B A schematic diagram schematically illustrates a process of calling a BERT instance to obtain a relevance score of a rewritten query relative to a query query according to an embodiment of the present disclosure.

[0089] like Figure 4B As shown, before BERT-based instance processing, the query and each rewritten query's combined query token IDs (the token ID sequence of the second string in the combined query), segment, position, etc. can be used as input to the second embedding 420 in combination with the second rewriting process, outputting a second information feature vector. During BERT-based instance processing, the second information feature vector can be used as input to the second encoder 421. Combined with the self-attention calculation and feedforward network processing in the second encoder 421, the relevance score for each rewriting type can be output, such as a synonymous rewriting score, a lower-level rewriting score, a higher-level rewriting score, a rewriting error score, etc., but is not limited to these.

[0090] For example, a BERT instance can calculate the relevance score and rewrite type for the query and each rewritten query based on the input second information feature vector. The BERT instance's output can include multiple rewrite types for each rewritten query and the corresponding relevance score for each rewritten type. The rewrite type with the highest relevance score can be used as the rewrite type and relevance value for the rewritten query. Only rewritten queries with relevance scores exceeding a set threshold can be distributed.

[0091] The above embodiments of the present disclosure are helpful in further improving the accuracy of the determined target correlation value.

[0092] Figure 5 The overall flow chart of online real-time generation of rewritten queries according to an embodiment of the present disclosure is schematically shown.

[0093] like Figure 5 As shown, the method includes operations S501 to S508.

[0094] In operation S501 , a query query is input.

[0095] In operation S502, a first feature extraction is performed on the query query to obtain a first information feature vector including a token id, a position, etc. representing a first character string in the query query.

[0096] In operation S503, the first information feature vector is input into a BART generation model, and a plurality of token ID sequences representing the rewritten character strings in the rewritten query are output.

[0097] In operation S504 , based on the token ID sequence representing the second character string in the rewritten query, a vocabulary is searched, and duplicate removal, black word filtering, and other processes are performed to generate a rewritten query.

[0098] In operation S505, a second feature extraction is performed on the rewritten query to obtain a second information feature vector including a token id, a segment, a position, etc. of a second character string in a combination of the query and the rewritten query.

[0099] In operation S506, the second information feature vector is input into the BERT discriminant model, and a rewriting type and a relevance value representing the rewritten query relative to the query are output.

[0100] In operation S507 , threshold filtering is performed based on the correlation value to determine the target rewriting query to be distributed.

[0101] In operation S508 , the distribution target rewrites the query.

[0102] To meet the time requirements of online deployment, the BART generative model and the BERT discriminative model can be deployed in a single service's inference engine to perform query rewriting. In a specific embodiment, generating 15 rewritten queries in real time for each search query can be completed within 20ms.

[0103] According to an embodiment of the present disclosure, relevant model parameters in the generative model and the discriminative model may use FP16 half-precision parameters.

[0104] Through the above-mentioned embodiments of the present disclosure, deploying the BART generation model and the BERT discrimination model on the same machine can reduce the RPC communication time consumption caused by the independent deployment of the two models. In order to maximize the reasoning performance, a high-performance reasoning engine can be developed based on the industry's open source reasoning framework. The bottleneck of reasoning performance can be analyzed in combination with the model structure. The use of FP16 half precision can further improve the reasoning speed. Compared with closed-source reasoning engines, it has better performance, flexibility and scalability.

[0105] Figure 6 The block diagram schematically shows an information distribution device according to an embodiment of the present disclosure.

[0106] like Figure 6 As shown, the information distribution device includes a first feature extraction module 610 , a rewriting module 620 , a correlation analysis module 630 and an information distribution module 640 .

[0107] The first feature extraction module 610 is configured to perform first feature extraction on the information query statement in response to the information query statement to obtain a first information feature vector.

[0108] The rewriting module 620 is used to schedule at least one generation model instance from the generation model pool, rewrite the first information feature vector, and obtain at least one information rewriting statement corresponding to the information query statement.

[0109] The correlation analysis module 630 is used to schedule at least one discriminant model instance from the discriminant model pool, perform correlation analysis on the information query statement and the information rewriting statement, and obtain the target rewriting statement whose correlation meets the preset conditions, wherein the generation model pool and the discriminant model pool are deployed in the same physical machine.

[0110] The information distribution module 640 is used to distribute information according to the target rewriting statement.

[0111] According to an embodiment of the present disclosure, the first feature extraction module includes a first tokenization processing unit, a first vectorization processing unit and a summation unit.

[0112] The first tokenization processing unit is configured to perform a first tokenization process on the information query statement to obtain a first character string sequence determined for the information query statement.

[0113] The first vectorization processing unit is configured to perform a first vectorization process on the first character string sequence to obtain a first character string identification feature sequence and a first character string position feature sequence of the first character string sequence.

[0114] The summing unit is configured to use a first summing operator to perform a sum calculation on the first character string identification feature sequence and the first character string position feature sequence to obtain a first information feature vector.

[0115] According to an embodiment of the present disclosure, the rewriting module includes a first encoding processing unit, a rewriting character string sequence obtaining unit, and an information rewriting sentence generating unit.

[0116] The first encoding processing unit is configured to perform a first encoding process on the first information feature vector to obtain at least one rewritten character string identifier sequence corresponding to the original character string sequence of the information query statement.

[0117] The rewritten character string sequence obtaining unit is configured to query an existing vocabulary table according to at least one rewritten character string identifier sequence to obtain at least one rewritten character string sequence, wherein a one-to-one mapping relationship between rewritten character string identifiers and rewritten characters is predefined in the existing vocabulary table.

[0118] The information rewriting statement generating unit is configured to generate at least one information rewriting statement according to at least one rewriting character string sequence.

[0119] According to an embodiment of the present disclosure, the information rewriting sentence generating unit includes a deduplication subunit, a deletion subunit, and a combination subunit.

[0120] The deduplication subunit is used to dedupe the rewritten character string sequence to obtain a first character string sequence.

[0121] The deletion subunit is configured to compare the first character string sequence with the blacklist character string, delete the target character string in the first character string sequence that matches the blacklist character string, and obtain a second character string sequence.

[0122] The combining subunit is used to combine the second character string sequence to generate an information rewriting statement.

[0123] According to an embodiment of the present disclosure, the correlation analysis module includes a second feature extraction unit, a second encoding processing unit, and a target rewriting sentence determination unit.

[0124] The second feature extraction unit is used to perform second feature extraction on the combination of the information query statement and any one of the information rewriting statements to obtain a second information feature vector.

[0125] The second encoding processing unit is used to perform a second encoding process on the second information feature vector to obtain a target correlation value between the information query statement and the information rewriting statement.

[0126] The target rewriting sentence determining unit is configured to determine an information rewriting sentence whose target relevance value satisfies a preset condition as a target rewriting sentence.

[0127] According to an embodiment of the present disclosure, the second feature extraction unit includes a second tokenization processing subunit, a second vectorization processing subunit and a summation subunit.

[0128] The second tokenization processing subunit is configured to perform a second tokenization process on a combination of the information query statement and any one of the information rewriting statements to obtain a second character string sequence determined for the combination statement.

[0129] The second vectorization processing subunit is configured to perform a second vectorization processing on the second character string sequence to obtain a second character string identification feature sequence and a second character string position feature sequence of the second character string sequence.

[0130] The summing subunit is configured to use a second summing operator to perform a sum calculation on the second character string identification feature sequence and the second character string position feature sequence to obtain a second information feature vector.

[0131] According to an embodiment of the present disclosure, the second encoding processing unit includes a second encoding processing sub-unit and a target correlation value determination sub-unit.

[0132] The second encoding processing subunit is used to perform second encoding processing on the second information feature vector, generate multiple rewriting types of the information rewriting statement relative to the information query statement and candidate relevance values ​​corresponding to each rewriting type, and obtain multiple candidate relevance values.

[0133] The target correlation value determination subunit is configured to determine the candidate correlation value with the largest value among the multiple candidate correlation values ​​as the target correlation value.

[0134] According to the embodiments of the present invention, any number of modules, units, and sub-units, or at least part of the functions of any number of them, can be implemented in one module. According to the embodiments of the present invention, any one or more of the modules, units, and sub-units can be split into multiple modules for implementation. According to the embodiments of the present invention, any one or more of the modules, units, and sub-units can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging the circuit, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or in an appropriate combination of any of them. Alternatively, according to the embodiments of the present invention, one or more of the modules, units, and sub-units can be at least partially implemented as a computer program module, which can perform the corresponding functions when the computer program module is executed.

[0135] For example, any number of the first feature extraction module 610, the rewriting module 620, the correlation analysis module 630, and the information distribution module 640 can be combined into a single module / unit / sub-unit, or any one of these modules / units / sub-units can be split into multiple modules / units / sub-units. Alternatively, at least part of the functionality of one or more of these modules / units / sub-units can be combined with at least part of the functionality of other modules / units / sub-units and implemented in a single module / unit / sub-unit. According to an embodiment of the present disclosure, at least one of the first feature extraction module 610, the rewriting module 620, the correlation analysis module 630, and the information distribution module 640 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware by any other reasonable means of circuit integration or packaging, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or any appropriate combination of any of these. Alternatively, at least one of the first feature extraction module 610 , the rewriting module 620 , the correlation analysis module 630 and the information distribution module 640 may be at least partially implemented as a computer program module, which may perform corresponding functions when executed.

[0136] It should be noted that the information distribution device part in the embodiment of the present disclosure corresponds to the information distribution method part in the embodiment of the present disclosure. The description of the information distribution device part specifically refers to the information distribution method part and will not be repeated here.

[0137] Figure 7 A block diagram schematically shows an electronic device suitable for implementing the information distribution method according to an embodiment of the present disclosure. Figure 7 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0138] like Figure 7 As shown, the electronic device 700 according to an embodiment of the present disclosure includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0139] Various programs and data required for the operation of the electronic device 700 are stored in the RAM 703. The processor 701, ROM 702, and RAM 703 are connected to each other via a bus 704. The processor 701 executes the various operations of the method flow according to the embodiment of the present disclosure by executing the programs in the ROM 702 and / or RAM 703. It should be noted that the programs may also be stored in one or more memories other than the ROM 702 and RAM 703. The processor 701 may also execute the various operations of the method flow according to the embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0140] According to an embodiment of the present disclosure, electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to bus 704. Electronic device 700 may also include one or more of the following components connected to I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 708 including a hard disk; and a communication section 709 including a network interface card such as a LAN card or modem. Communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to I / O interface 705 as needed. Removable media 711, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 710 as needed, so that computer programs read from the removable media can be installed into storage section 708 as needed.

[0141] According to an embodiment of the present disclosure, the method flow according to an embodiment of the present disclosure can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, the above-mentioned functions defined in the system of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0142] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when executed, implements the method according to the embodiments of the present disclosure.

[0143] According to embodiments of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium. Examples include, but are not limited to, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0144] For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 702 and / or the RAM 703 described above and / or one or more memories other than the ROM 702 and the RAM 703 .

[0145] An embodiment of the present disclosure also includes a computer program product, which includes a computer program, which contains program code for executing the method provided by the embodiment of the present disclosure. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the information distribution method provided by the embodiment of the present disclosure.

[0146] When the computer program is executed by the processor 701, the above functions defined in the system / device of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0147] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 709, and / or installed from a removable medium 711. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0148] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0149] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, which contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, as well as the combination of boxes in the block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or operations, or can be implemented using a combination of dedicated hardware and computer instructions. It will be understood by those skilled in the art that the features described in the various embodiments of the present disclosure can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways, and all of these combinations and / or couplings fall within the scope of the present disclosure.

[0150] The above describes the embodiments of the present disclosure. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be advantageously used in combination. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. An information distribution method, comprising: In response to an information query statement, performing a first feature extraction on the information query statement to obtain a first information feature vector; Dispatching at least one generative model instance from a generative model pool, rewriting the first information feature vector to obtain at least one information rewriting statement corresponding to the information query statement; Dispatching at least one discriminant model instance from a discriminant model pool, performing a correlation analysis on the information query statement and the information rewriting statement, and obtaining a target rewriting statement whose correlation meets a preset condition, wherein the generation model pool and the discriminant model pool are deployed in the same physical machine; as well as The sentence is rewritten according to the target and information is distributed.

2. The method according to claim 1, wherein The performing a first feature extraction on the information query statement to obtain a first information feature vector includes: performing a first tokenization process on the information query statement to obtain a first character string sequence determined for the information query statement; performing a first vectorization process on the first character string sequence to obtain a first character string identification feature sequence and a first character string position feature sequence of the first character string sequence; and A first summation operator is used to perform a summation calculation on the first character string identification feature sequence and the first character string position feature sequence to obtain the first information feature vector.

3. The method according to claim 1, wherein The rewriting of the first information feature vector to obtain at least one information rewriting statement corresponding to the information query statement includes: performing a first encoding process on the first information feature vector to obtain at least one rewritten character string identifier sequence corresponding to the original character string sequence of the information query statement; According to at least one of the rewritten character string identifier sequences, querying an existing vocabulary table to obtain at least one rewritten character string sequence, wherein the existing vocabulary table predefines a one-to-one mapping relationship between rewritten character string identifiers and rewritten character strings; and At least one of the information rewriting sentences is generated according to at least one of the rewriting character string sequences.

4. The method according to claim 3, wherein: Generating at least one of the information rewriting statements according to at least one of the rewriting character string sequences includes: Deduplicating the rewritten character string sequence to obtain a first character string sequence; Comparing the first character string sequence with a blacklist character string, deleting a target character string in the first character string sequence that matches the blacklist character string, to obtain a second character string sequence; and The second character string sequence is combined to generate the information rewriting statement.

5. The method according to claim 1, wherein The performing of correlation analysis on the information query statement and the information rewriting statement to obtain a target rewriting statement whose correlation meets a preset condition comprises: Performing second feature extraction on a combination of the information query statement and any one of the information rewriting statements to obtain a second information feature vector; performing a second encoding process on the second information feature vector to obtain a target correlation value between the information query statement and the information rewriting statement; and The information rewriting sentence whose target relevance value satisfies the preset condition is determined as the target rewriting sentence.

6. The method according to claim 5, wherein: The performing second feature extraction on the combined sentence of the information query sentence and any one of the information rewriting sentences to obtain the second information feature vector comprises: performing a second tokenization process on a combination of the information query statement and any one of the information rewriting statements to obtain a second character string sequence determined for the combination statement; performing a second vectorization process on the second character string sequence to obtain a second character string identification feature sequence and a second character string position feature sequence of the second character string sequence; and The second character string identification feature sequence and the second character string position feature sequence are summed using a second summation operator to obtain the second information feature vector.

7. The method according to claim 5 or 6, wherein: The performing a second encoding process on the second information feature vector to obtain a target correlation value between the information query statement and the information rewriting statement includes: performing a second encoding process on the second information feature vector to generate a plurality of rewriting types of the information rewriting statement relative to the information query statement and a candidate relevance value corresponding to each of the rewriting types, thereby obtaining a plurality of candidate relevance values; and The candidate correlation value with the largest value among the multiple candidate correlation values ​​is determined as the target correlation value.

8. An information distribution device, comprising: A first feature extraction module is configured to extract a first feature of the information query statement in response to the information query statement to obtain a first information feature vector; a rewriting module, configured to schedule at least one generative model instance from a generative model pool, rewrite the first information feature vector, and obtain at least one information rewriting statement corresponding to the information query statement; a correlation analysis module, configured to schedule at least one discriminant model instance from a discriminant model pool, perform correlation analysis on the information query statement and the information rewriting statement, and obtain a target rewriting statement whose correlation satisfies a preset condition, wherein the generation model pool and the discriminant model pool are deployed on the same physical machine; as well as The information distribution module is used to distribute information according to the target rewriting statement.

9. An electronic device comprising: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to implement the method according to any one of claims 1 to 7.

11. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images