Data processing method, apparatus and electronic device

Abstract

The present disclosure provides a data processing method and device and electronic equipment, input the obtained user question description information and user input information into a data processing model to obtain an output vector; select a target data vector matching the output vector from a preset question and answer library, determine the question and answer reply information corresponding to the target data vector in the preset question and answer library; the preset question and answer library includes a plurality of question and answer pairs and data vectors corresponding to the question and answer pairs; the question and answer pair includes question description information and question and answer reply information; based on the question and answer reply information corresponding to the target data vector, generate target reply information corresponding to the user question description information. In this way, based on the user question description information and the user input information, more accurate question and answer records can be recalled from the preset question and answer library, and question and answer reply information is generated based on the question and answer records, so that the user's intention can be accurately understood and the correct reply can be given, and the customer service question reply efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing design, and in particular, to a data processing method, device and electronic equipment. BACKGROUND

[0002] Under the background of rapid development of Internet technology, the customer consultation volume faced by enterprises presents an explosive growth trend. This growth poses a huge challenge to the traditional manual customer service mode, which is limited by human resources and response speed and is difficult to meet the increasingly high customer service demand.

[0003] In the related art, a manner of replying to user questions based on a large language model is provided. Before replying to a user question, the manner searches for a question related to the user question from a product-prepared question and answer knowledge base, and then provides the user question to the large language model to generate an answer related to the product. However, relying solely on the semantic information of the user question sometimes cannot search for accurate records, and the user question and the records stored in the question and answer knowledge base may have a logical relationship. In addition, even if the same question is described, the reply given may be different due to the different states of the user. SUMMARY

[0004] The purpose of the present disclosure is to provide a data processing method, device and electronic equipment to effectively improve the reply efficiency of customer service, thereby reducing the operating cost on the basis of providing timely replies to users.

[0005] In a first aspect, the present disclosure provides a data processing method, which comprises: obtaining user question description information and user input information; wherein the user input information is used to indicate the information sending state and log information corresponding to the user question description information; inputting the user question description information and / or the user input information into a preset data processing model trained to obtain an output vector; selecting a target data vector matching the output vector from a preset question and answer library to determine the question reply information corresponding to the target data vector in the preset question and answer library; wherein the preset question and answer library comprises a plurality of question and reply pairs and data vectors corresponding to the question and reply pairs; the question and reply pair comprises question description information and question reply information corresponding to the question description information; and generating target reply information corresponding to the user question description information based on the question reply information corresponding to the target data vector.

[0006] In a second aspect, the present disclosure provides a data processing apparatus, comprising: an information acquisition module configured to acquire user problem description information and user input information; wherein the user input information is used to indicate information sending state and / or log information corresponding to the user problem description information; a vector determination module configured to input the user problem description information and the user input information into a preset data processing model trained, to obtain an output vector; a reply determination module configured to select a target data vector matching the output vector from a preset question and answer library, and determine problem reply information corresponding to the target data vector in the preset question and answer library; wherein the preset question and answer library comprises a plurality of problem and reply pairs, and data vectors corresponding to the problem and reply pairs; the problem and reply pair comprises problem description information and problem reply information corresponding to the problem description information; and a reply generation module configured to generate target reply information corresponding to the user problem description information based on the problem reply information corresponding to the target data vector.

[0007] In a third aspect, the present disclosure provides an electronic device, comprising a processor and a memory, wherein the memory stores machine executable instructions executable by the processor, and the processor executes the machine executable instructions to implement the data processing method.

[0008] In a fourth aspect, the present disclosure provides a computer readable storage medium, wherein the computer readable storage medium stores computer executable instructions, and when the computer executable instructions are invoked and executed by a processor, the computer executable instructions cause the processor to implement the data processing method.

[0009] The embodiments of the present disclosure bring the following beneficial effects:

[0010] The data processing method, apparatus and electronic device provided by the present disclosure first acquire user problem description information and user input information; wherein the user input information is used to indicate information sending state and / or log information corresponding to the user problem description information; then input the user problem description information and the user input information into a preset data processing model trained, to obtain an output vector; further select a target data vector matching the output vector from a preset question and answer library, and determine problem reply information corresponding to the target data vector in the preset question and answer library; wherein the preset question and answer library comprises a plurality of problem and reply pairs, and data vectors corresponding to the problem and reply pairs; the problem and reply pair comprises problem description information and problem reply information corresponding to the problem description information; and then generate target reply information corresponding to the user problem description information based on the problem reply information corresponding to the target data vector. In this way, based on the user problem description information and the user input information, more accurate question and answer records can be recalled from the preset question and answer library, and problem reply information is generated based on the question and answer records, so that the user's intention can be accurately understood and correct replies can be given, and the efficiency of replying to customer service problems can be improved.

[0011] Other features and advantages of the present disclosure will be set forth in the descriptions that follow, and in part will be apparent from the description, or can be learned by practice of the present disclosure as hereinafter more fully described, or can be learned by practice of the present disclosure.

[0012] In order to make the above objectives, features and advantages of the present disclosure more obvious and comprehensible, the following preferred embodiments are specifically described below, and the accompanying drawings are referred to for a detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present disclosure, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0014] Figure 1 A flow chart of a data processing method provided by an embodiment of the present disclosure;

[0015] Figure 2 An architectural diagram of generating a recommended reply provided by an embodiment of the present disclosure;

[0016] Figure 3 A structural schematic diagram of a data processing apparatus provided by an embodiment of the present disclosure;

[0017] Figure 4 A structural schematic diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] In order to make the objectives, technical solutions and advantages of the embodiments of the present disclosure more clear, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only some of the embodiments of the present disclosure, but not all the embodiments. The components of the embodiments of the present disclosure described and shown in the drawings herein can be arranged and designed in various different configurations.

[0019] Therefore, the following detailed description of the embodiments of the present disclosure provided in the drawings is not intended to limit the scope of the claimed present disclosure, but only represents selected embodiments of the present disclosure. All other embodiments obtained by those skilled in the art without creative labor on the basis of the embodiments in the present disclosure belong to the scope of protection of the present disclosure.

[0020] In the context of the rapid development of Internet technology, the number of customer inquiries faced by enterprises is showing an explosive growth trend. This growth poses a huge challenge to the traditional manual customer service mode, which is limited by human resources and response speed and is difficult to meet the increasingly high customer service demand. To address this challenge, intelligent customer service systems have emerged and gradually become a key tool for enterprises to improve service efficiency and reduce operating costs.

[0021] The feedback system in the related art is the traditional customer service mode, in which users of each product submit problem tickets on their own, and customer service personnel process and reply to user problems one by one. Limited by human resources, when a large number of problem tickets are encountered, the response speed is greatly reduced, and the user experience is poor. Since large language models (LLMs, such as chatgpt, etc.) have made a breakthrough in natural language understanding and processing capabilities, they can effectively understand user intent and generate more appropriate and natural replies, and therefore have also been widely applied to customer service scenarios to improve reply efficiency. However, due to the "hallucination" problem of LLMs and the fact that many user problems are strongly related to the product, it is not feasible to directly use LLMs to reply to user problems. To this end, a retrieval-augmented generation (RAG) technology has emerged, the core idea of which is to retrieve relevant problems from a product-prepared question-and-answer knowledge base before replying to a user problem, and then provide the user problem and the relevant FAQs as context to the LLM to generate more accurate and product-related answers.

[0022] However, in actual application, this process still has two problems:

[0023] 1. The product needs to prepare a question-and-answer knowledge base in advance, and due to timeliness and product improvement, the knowledge base needs to be maintained and updated to ensure the reply effect, which brings certain maintenance costs

[0024] 2. The current RAG technology retrieves relevant problems from the question-and-answer knowledge base based on user problems, which is basically based on a semantic vector model, and the semantic relevance of the user problem and each record in the question-and-answer knowledge base is calculated to finally filter the results. However, in actual application, user problems are diverse, and relying solely on semantic information sometimes cannot retrieve accurate records. In addition, even for the same problem description, the reply given may be different due to the user's own state.

[0025] Based on the above problems, the embodiments of the present disclosure provide a data processing method, device, and electronic equipment, which can be applied to a problem reply scenario.

[0026] To facilitate understanding of the embodiments of this disclosure, a data processing method disclosed in this disclosure will first be described in detail, such as... Figure 1 As shown, the method includes the following specific steps:

[0027] Step S102: Obtain user problem description information and user input information; wherein, the user input information is used to indicate the information sending status and / or log information corresponding to the user problem description information.

[0028] The aforementioned user problem description information is the text of a problem submitted by the user through their device, which can be a mobile phone, computer, or tablet. The aforementioned user input information may include one or more of the following: user status information, user behavior information, and recorded log information; wherein, user status information indicates the device status when the user sends the user problem description information, user behavior information indicates the actions performed by the user on their device when or prior to sending the user problem description information, and log information may record information such as errors or vulnerabilities.

[0029] In practice, when a user submits a problem description through their device, they also submit user input information so that a response can be provided based on both the problem description and the user input.

[0030] Step S104: Input the user's problem description information and user input information into the preset trained data processing model to obtain the output vector.

[0031] The data processing model described above can be a neural network model or a deep learning model, etc., and the specific model can be determined according to the research and development needs. This data processing model is pre-trained, and the training method can be determined according to the research and development needs. The function of this data processing model is to analyze the input data, convert the input data into an embedding vector, and output the embedding vector, which is the aforementioned output vector.

[0032] Step S106: Select a target data vector that matches the output vector from the preset question-and-answer database, and determine the question-and-answer information corresponding to the target data vector in the preset question-and-answer database; wherein, the preset question-and-answer database includes multiple question-and-answer pairs and the data vectors corresponding to the question-and-answer pairs; the question-and-answer pairs include question description information and question-and-answer information corresponding to the question description information.

[0033] In a specific implementation, the preset question and answer library includes a large number of question and answer pairs, and each question and answer pair corresponds to a data vector; each question and answer pair includes question description information and question and answer information corresponding to the question description information. After obtaining the output vector corresponding to the user question description information, the similarity between the output vector and the data vector corresponding to each question and answer pair in the preset question and answer library needs to be calculated, and then one or more data vectors with high similarity to the output vector are selected from the preset question and answer library as target data vectors based on a preset rule, and the target question and answer pair corresponding to the target data vector is determined, and the question and answer information in the target question and answer pair is determined as the question and answer information corresponding to the target data vector.

[0034] The preset rule can be determined according to research and development requirements. For example, the preset rule can be to select one data vector with the highest similarity to the output vector from the preset question and answer library as the target data vector, or to select multiple data vectors with high similarity to the output vector from the preset question and answer library as the target data vector, or to select multiple data vectors of different categories with high similarity to the output vector from the preset question and answer library as the target data vector, and the like.

[0035] In the related art, relevant records are retrieved from the preset question and answer library based on semantic vectors, but due to the diversity of user question description information, the same question description information may correspond to different reply records in different states, and the like, and simple semantic matching cannot handle these problems. Based on the user question description information and the user input information, the disclosure can recall some more accurate question and answer records from the preset question and answer library.

[0036] In step S108, target reply information corresponding to the user question description information is generated based on the question and answer information corresponding to the target data vector.

[0037] In a specific implementation, if the question and answer information corresponding to the target data vector is one, the question and answer information corresponding to the target data vector can be determined as the target reply information corresponding to the user question description information; if the question and answer information corresponding to the target data vector is one or more, the target data vector needs to be integrated and analyzed to obtain the target reply information corresponding to the user question description information. Specifically, the question and answer information corresponding to the target data vector can be input into a pre-trained large language model to obtain the target reply information corresponding to the user question description information; or the question and answer information corresponding to the target data vector can be analyzed by voice to obtain a sentence that is smooth.

[0038] The data processing method provided by the embodiment of the present disclosure can recall more accurate question and answer records from the preset question and answer library based on the user question description information and the user input information, and generate question reply information based on the question and answer records, so that the user's intention can be accurately understood and a correct reply can be given, and the efficiency of replying to customer service questions can be improved.

[0039] The following embodiments are used to describe the manner of determining the target data vector and the manner of determining the preset question and answer library.

[0040] Specifically, the specific process of selecting the target data vector matching the output vector from the preset question and answer library can include: determining the similarity of the output vector and the data vector corresponding to each question and answer pair in the preset question and answer library; and selecting the target data vector matching the output vector from the preset question and answer library based on the similarity.

[0041] In a specific implementation, when determining the target data vector, the similarity of the output vector and the data vector corresponding to each question and answer pair in the preset question and answer library is first calculated to obtain the similarity corresponding to each data vector, and then one or more data vectors with higher similarity are selected from the preset question and answer library as the target data vector.

[0042] In an optional embodiment, the specific process of selecting the target data vector matching the output vector from the preset question and answer library can be performed by any one of the following methods:

[0043] Method one: the data vectors in the preset question and answer library with a similarity greater than a first preset similarity threshold value are determined as the target data vectors matching the output vector.

[0044] In a specific implementation, the first preset similarity threshold value can be determined according to research and development requirements, for example, the first preset similarity threshold value can be 85% or 90%, etc. Specifically, all data vectors in the preset question and answer library with a similarity greater than the first preset similarity threshold value can be determined as the target data vectors.

[0045] Method two: the similarities of the output vector and the data vectors corresponding to the question and answer pairs in the preset question and answer library are sorted in descending order to obtain a sorting result, and a specified number of data vectors at the front of the sorting result are determined as the target data vectors matching the output vector.

[0046] In a specific implementation, first, the similarity corresponding to each data vector in the preset question and answer library can be sorted in descending order of similarity to obtain a sorting result, and then a specified number of data vectors at the front of the sorting result are selected as the target data vectors, and the specified number can be determined according to research and development requirements. For example, the specified number can be 5 or 6, etc.

[0047] In another optional embodiment, the specific process of selecting the target data vector matching the output vector from the preset question and answer library based on the similarity can further include: selecting a first data vector with the highest similarity to the output vector from each cluster group in the preset question and answer library; determining the first data vector in each cluster group as the target data vector matching the output vector; and wherein the number of vectors contained in the target data vector matches the number of cluster groups contained in the preset question and answer library.

[0048] In specific implementation, the preset question and answer library contains a plurality of cluster groups, each cluster group contains one or more question and answer pairs, and each question and answer pair corresponds to a data vector. At the same time, the similarity between the data vectors contained in each cluster group is high. Therefore, when determining the target data vector, one data vector with the highest similarity to the output vector can be selected from each cluster group as the target data vector, so that the number of vectors contained in the target data vector is the same as the number of cluster groups contained in the preset question and answer library.

[0049] In actual application, the preset question and answer library is determined by: obtaining a plurality of historical reply records; wherein the historical reply record includes a question and answer pair composed of question description information and question and answer information; for each historical reply record in the plurality of historical reply records, inputting the historical reply record into a data processing model to obtain a data vector corresponding to the historical reply record; determining the similarity between the data vectors corresponding to each two historical reply records in the plurality of historical reply records; dividing the historical reply records with a similarity less than a second preset threshold into the same cluster group to obtain a plurality of cluster groups; wherein the cluster group includes at least one of the plurality of historical reply records and the data vectors corresponding to the plurality of historical reply records; and determining the plurality of cluster groups as the preset question and answer library.

[0050] In specific implementation, when initially constructing the preset question and answer library, first, the historical reply records of user questions (which are question and answer pairs) need to be obtained, and then the data processing model is used to convert the batch of historical reply records into data vectors, i.e., each historical reply record corresponds to a data vector. After obtaining all the data vectors, the data vectors and historical reply records are clustered, i.e., similar historical reply records are clustered into a group. If the similarity between a historical reply record and the clustered data is lower than the second preset similarity threshold, a new cluster group is added. Finally, the obtained historical reply records are converted into N cluster groups, each cluster group includes at least one historical reply record and the data vector corresponding to the historical reply record. The initial historical question and answer library is the N cluster groups.

[0051] In order to ensure the accuracy of the recommended reply, the comprehensiveness and real-time of the preset question and answer library is very important, but relying on each product person to maintain the preset question and answer library every day will waste a lot of manpower. Therefore, the present disclosure provides a way to update the preset question and answer library. Specifically, the way to update the preset question and answer library includes: grouping the user question description information and the target reply information into a target question and reply pair, and determining the output vector as the target data vector corresponding to the target question and reply pair; based on the similarity between the target data vector and the data vector corresponding to each question and reply pair in the preset question and answer library, determine the target clustering group to which the target data vector belongs in the preset question and answer library; store the target question and reply pair and the target data vector in the target clustering group in association to update the preset question and answer library.

[0052] In specific implementation, whenever the customer service replies a new user question, that is, each time the target reply information corresponding to the user question description information sent by the user is obtained, the output vector corresponding to the user question description information is obtained or the user question description information and the target reply information are converted into the output vector again, and then the similarity between the output vector and each data vector in the preset question and answer library is calculated. If the similarity between the output vector and a certain data vector is greater than a second preset similarity threshold, the target question and reply pair corresponding to the output vector is divided into the clustering group corresponding to the data vector; if the similarity between the output vector and each data vector in the preset question and answer library is not greater than the second preset similarity threshold, the output vector and the corresponding target question and reply are divided into a new clustering group, so as to ensure that the uncovered hot issues can also be effectively handled.

[0053] The following embodiments are used to describe the way to generate the target reply information.

[0054] Specifically, the specific process of generating the target reply information corresponding to the user question description information based on the question and reply information corresponding to the target data vector can include: determining the model prompt word based on the user question description information, the question and reply information corresponding to the target data vector, and the information output rule; inputting the model prompt word into the large language model which has been pre-trained to obtain the target reply information corresponding to the user question description information.

[0055] In a specific implementation, the large language model mentioned above refers to a natural language processing model trained using a large amount of data, which can understand natural language and generate human-readable text. The large language model used in the embodiments of the present disclosure is pre-trained, and through learning and analysis of a large amount of data, the large language model can automatically generate text with certain logical structure and semantics. Specifically, when using the large language model, how to design the input prompt word is very important, because the output of the large language model (i.e., the generated text) will be generated according to the input prompt word. Good prompt words can guide the large language model to generate useful and accurate text, while poor prompt words can cause the output of the large language model to deviate from expectations, or produce ambiguous and unclear results.

[0056] In a specific implementation, the model prompt word contains user question description information, target data vector corresponding question reply information, and information output rules. For example, the model prompt word can be: "You are a customer service of an xxx product, and you are required to give a reply. Here, the user question is the user description information, and there are some historical reply references: target data vector corresponding question reply information. Please output the target reply information according to the information output rules yyy." Specifically, the information output rules mentioned above can be determined according to the research and development requirements, which are not limited here.

[0057] The following embodiments are used to describe the way of translating data.

[0058] Specifically, after obtaining the user question description information and the user input information, the first language corresponding to the user question description information can be identified; then the user question description information and the user input information are input into the preset trained data processing model to obtain an output vector; a target data vector matching the output vector is selected from the preset question and answer library, the question reply information corresponding to the target data vector in the preset question and answer library is determined, and based on the question reply information corresponding to the target data vector, the target reply information corresponding to the user question description information is generated. After generating the target reply information corresponding to the user question description information based on the question reply information corresponding to the target data vector, the second language corresponding to the target reply information corresponding to the user question description information is identified; it is determined whether the first language is the same as the second language; if so, the target reply information is sent to the user equipment sending the user question description information; if not, the target reply information is translated into the first language, and the target reply information in the first language is returned to the user equipment sending the user question description information.

[0059] In actual application scenarios, the problem of multiple languages is often encountered. For example, the user may ask questions in English, Korean, or some small language, and obtain user question description information. Therefore, the recommended reply also needs to add a translation step to give a reply in the corresponding language based on the language of the user's question. Specifically, first, it is necessary to identify which country language the user submitted user question description information is in. Here, the CompactLanguage Detector language recognition method based on machine learning can be used to identify the first language to be translated, which is the language corresponding to the user question description information. If the obtained target reply information is also in the first language, no translation is needed, otherwise the second language corresponding to the target reply information needs to be translated into the first language.

[0060] In a specific implementation, the specific process of translating the second language corresponding to the target reply information into the first language can include: first, the target term words matching the term words in the term table need to be extracted from the target reply information, and then a set of prompt words is constructed by combining the extracted second language target reply information, target term words and first language information, etc. The prompt words are input into the translation model to obtain the target reply information in the first language. The term table includes a plurality of term words, and each term word corresponds to a translated text in different languages.

[0061] In a specific embodiment, the above prompt words can be: "You are a language expert, and you are required to translate from the second language to the first language. The sentence to be translated is the target reply information in the first language. Here are some translation term words: target data words in the first language. Please output the translation result." The above translation model can also be a large language model.

[0062] The following embodiments are used to describe the classification and label screening method of the user question.

[0063] Specifically, after obtaining the user question description information and the user input information, it is determined whether the user question description information is valid information. If it is valid information, the step of inputting the user question description information and the user input information into the preset trained data processing model is performed. If it is not valid information, the preset reply information is returned to the user device that sent the user question description information.

[0064] In a specific implementation, for the user-submitted user question description information, it is first needed to determine whether it is a valid question. In actual scenarios, invalid questions submitted by users are often received, and if these questions all need to be reviewed by customer service personnel one by one, too much time is wasted. Therefore, the present disclosure designs an effectiveness analysis module, which can analyze the submitted question by using an NLP text classification technology. For example, if the question text description information submitted by the user does not have enough semantic information or is a jumbled description (for example, 111111, asdfgg, etc.), it can be determined that the question description information is invalid. For the question description information determined to be invalid, a preset reply information can be given immediately to prompt the user to supplement information to perfect the question, so as to avoid the user waiting for too long and also reduce the intervention of customer service personnel.

[0065] In a specific embodiment, the manner of determining whether the user question description information is valid information can further include inputting the user question description information into a classification model, which can output whether the user question description information is valid. The classification model is trained based on training data, and the training data includes a large number of training samples, each of which includes question description information and a validity label annotated by an annotator for the question description information, the validity label being used to indicate that the question description information is valid or invalid. The above-mentioned classification model can be a Bert model or other neural network model, which can be determined according to research and development requirements.

[0066] In an optional embodiment, after obtaining the user question description information and the user input information, the user question description information is input into a pre-trained label model to obtain a target label corresponding to the user question description information; the target label is a label in a plurality of preset labels.

[0067] In a specific implementation, the above-mentioned label model is trained based on preset training data, and the preset training data includes a plurality of samples, each of which includes question description information and a category label annotated by an annotator for the question description information, the category label being any one of a plurality of preset labels, for example, the preset labels can include but are not limited to a regular question, an abnormal error, and a recharge question, etc. Specifically, for the user question description information determined to be a valid question, the label model needs to be used to further analyze it to determine the category (equivalent to the above-mentioned target label) to which the user question description information belongs. This step is mainly to determine the importance of the user question description information, and when customer service personnel simultaneously receive too many user question description information, the category can be used to filter and prioritize the response to more important questions, thereby bringing better user experience.

[0068] For example, Figure 2An architecture diagram of generating a recommended reply is shown, a user first inputs user question description information, then performs validity analysis, when it is determined that the user question description information is invalid question, preset reply information is output. When it is determined that the user question description information is valid question, a target label corresponding to the user question description information is identified to determine the processing priority of the user question description information; then the valid user question description information is input into a personalized retrieval model (equivalent to the above data processing model) to obtain an output vector corresponding to the user question description information, the similarity of the output vector and a data vector corresponding to each question and reply pair in a preset question and answer library is determined, K target data vectors with higher similarity to the output vector are selected from the preset question and answer library; then question and reply information corresponding to the target data vector in the preset question and answer library is determined, and a model prompt word is determined based on the user question description information, the question and reply information corresponding to the target data vector, and an information output rule; the model prompt word is input into a large language model (equivalent to the LLM model in Figure 2

[0069] By using the above technical solution, the reply efficiency of the customer service can be effectively improved, so as to reduce the operating cost on the basis of providing timely reply to the user.

[0070] Corresponding to the above method embodiment, the present embodiment also provides a data processing device, as shown in Figure 3

[0071] An information acquisition module 30 is configured to acquire user question description information and user input information; wherein the user input information is used to indicate the information sending state and / or log information corresponding to the user question description information.

[0072] A vector determination module 31 is configured to input the user question description information and the user input information into a preset trained data processing model to obtain an output vector.

[0073] A reply determination module 32 is configured to select a target data vector matching the output vector from a preset question and answer library, and determine question and reply information corresponding to the target data vector in the preset question and answer library; wherein the preset question and answer library comprises a plurality of question and reply pairs, and data vectors corresponding to the question and reply pairs; the question and reply pair comprises question description information and question and reply information corresponding to the question description information.

[0074] A reply generation module 33 is configured to generate target reply information corresponding to the user question description information based on the question and reply information corresponding to the target data vector.​​

[0075] The data processing apparatus can recall more accurate question and answer records from the preset question and answer library based on the user question description information and the user input information, and generate question reply information based on the question and answer records, so that the user's intention can be accurately understood and a correct reply can be given, and the efficiency of replying to customer questions can be improved.

[0076] Specifically, the reply determination module 32 is configured to: determine the similarity between the output vector and the data vector corresponding to each question and answer pair in the preset question and answer library; and select a target data vector matching the output vector from the preset question and answer library based on the similarity.

[0077] Further, the reply determination module 32 is configured to: determine the data vector with a similarity greater than a first preset similarity threshold in the preset question and answer library as the target data vector matching the output vector; or sort the similarity between the output vector and the data vector corresponding to each question and answer pair in the preset question and answer library in descending order to obtain a sorting result, and determine a specified number of data vectors at the front of the sorting result as the target data vector matching the output vector.

[0078] Further, the apparatus further includes a question and answer library determination module configured to: obtain a plurality of historical reply records; wherein the historical reply records include question and answer pairs composed of question description information and question reply information; input each historical reply record in the plurality of historical reply records into a data processing model to obtain a data vector corresponding to the historical reply record; determine the similarity between the data vectors corresponding to each two historical reply records in the plurality of historical reply records; divide the historical reply records with a similarity less than a second preset threshold into the same clustering group to obtain a plurality of clustering groups; wherein the clustering group includes at least one plurality of historical reply records and the data vectors corresponding to the plurality of historical reply records; and determine the plurality of clustering groups as the preset question and answer library.

[0079] Further, the apparatus further includes a question and answer library updating module configured to: compose a target question and answer pair with the user question description information and the target reply information, and determine a target data vector corresponding to the target question and answer pair; determine a target clustering group to which the target data vector belongs in the preset question and answer library based on the similarity between the target data vector and the data vector corresponding to each question and answer pair in the preset question and answer library; and store the target question and answer pair and the target data vector in the target clustering group in association to update the preset question and answer library.

[0080] Further, the reply determination module 32 is further configured to: select a first data vector with the highest similarity to the output vector from each cluster group in the preset question and answer library; and determine the first data vector in each cluster group as a target data vector matched with the output vector, wherein the target data vector contains a number of vectors matched with the number of cluster groups in the preset question and answer library.

[0081] Further, the reply generation module 33 is further configured to: determine a model prompt word based on the user question description information, the question and answer reply information corresponding to the target data vector, and the information output rule; and input the model prompt word into a pre-trained large language model to obtain the target reply information corresponding to the user question description information.

[0082] Further, the device further comprises an effectiveness determination module configured to: determine whether the user question description information is valid information after obtaining the user question description information and the user input information; if the user question description information is valid information, execute the step of inputting the user question description information and the user input information into the pre-trained data processing model; and if the user question description information is not valid information, return preset reply information to the user equipment that sends the user question description information.

[0083] Further, the device further comprises a label setting module configured to: input the user question description information into a pre-trained label model to obtain a target label corresponding to the user question description information after obtaining the user question description information and the user input information, wherein the target label is a label in a plurality of preset labels.

[0084] Further, the device further comprises a language recognition module configured to: recognize a first language corresponding to the user question description information after obtaining the user question description information and the user input information. Based on this, the device further comprises a translation module configured to: recognize a second language corresponding to the target reply information corresponding to the user question description information after generating the target reply information corresponding to the user question description information based on the question and answer reply information corresponding to the target data vector; determine whether the first language is the same as the second language; if the first language is the same as the second language, return the target reply information to the user equipment that sends the user question description information; and if the first language is not the same as the second language, translate the target reply information into the first language, and return the target reply information in the first language to the user equipment that sends the user question description information.

[0085] The data processing device provided in the embodiments of the present disclosure has the same implementation principle and technical effects as the foregoing method embodiments. For brevity of description, the part not mentioned in the device embodiment part can be referred to the corresponding content in the foregoing method embodiments.

[0086] The embodiments of the present disclosure also provide an electronic device, such as Figure 4As shown, the electronic device includes a processor and a memory having stored therein machine executable instructions executable by the processor to implement the above data processing method.

[0087] Specifically, the above data processing method comprises: obtaining user problem description information and user input information; wherein the user input information is used to indicate the information sending state and / or log information corresponding to the user problem description information; inputting the user problem description information and the user input information into a preset data processing model trained to obtain an output vector; selecting a target data vector matching the output vector from a preset question and answer library, and determining question and answer information corresponding to the target data vector in the preset question and answer library; wherein the preset question and answer library comprises a plurality of question and answer pairs and data vectors corresponding to the question and answer pairs; the question and answer pair comprises question description information and question answer information corresponding to the question description information; and generating target answer information corresponding to the user problem description information based on the question and answer information corresponding to the target data vector.

[0088] In the above data processing method, based on the user problem description information and the user input information, more accurate question and answer records can be recalled from the preset question and answer library, and question and answer information can be generated based on the question and answer records, so that the method can accurately understand the user's intention and give correct replies, and improve the efficiency of customer service problem replies.

[0089] In an optional embodiment, the step of selecting a target data vector matching the output vector from the preset question and answer library comprises: determining the similarity of the output vector and the data vector corresponding to each question and answer pair in the preset question and answer library; and selecting the target data vector matching the output vector from the preset question and answer library based on the similarity.

[0090] In an optional embodiment, the step of selecting a target data vector matching the output vector from the preset question and answer library based on the similarity comprises one of the following ways: determining the data vector in the preset question and answer library with a similarity greater than a first preset similarity threshold as the target data vector matching the output vector; and sorting the similarity of the output vector and the data vector corresponding to each question and answer pair in the preset question and answer library in descending order to obtain a sorting result, and determining the data vectors with a specified number of top sorting results as the target data vectors matching the output vector.

[0091] In an optional embodiment, the preset question and answer library is determined by: obtaining a plurality of historical reply records; wherein, the historical reply record comprises a question and answer pair composed of question description information and question reply information; for each historical reply record in the plurality of historical reply records, inputting the historical reply record into a data processing model to obtain a data vector corresponding to the historical reply record; determining the similarity between the data vectors corresponding to each two historical reply records in the plurality of historical reply records; dividing the historical reply records with a similarity less than a second preset threshold into the same cluster group to obtain a plurality of cluster groups; wherein, the cluster group comprises at least one plurality of historical reply records and the data vectors corresponding to the plurality of historical reply records; and determining the plurality of cluster groups as the preset question and answer library.

[0092] In an optional embodiment, the method further comprises: composing a target question and answer pair from the user question description information and the target reply information, and determining the output vector as a target data vector corresponding to the target question and answer pair; determining a target cluster group to which the target data vector belongs in the preset question and answer library based on the similarity between the target data vector and the data vector corresponding to each question and answer pair in the preset question and answer library; and storing the target question and answer pair and the target data vector in the target cluster group in association to update the preset question and answer library.

[0093] In an optional embodiment, the step of selecting a target data vector matching the output vector from the preset question and answer library based on the similarity comprises: selecting a first data vector with the highest similarity to the output vector from each cluster group in the preset question and answer library; and determining the first data vector in each cluster group as the target data vector matching the output vector; wherein, the number of vectors contained in the target data vector matches the number of cluster groups contained in the preset question and answer library.

[0094] In an optional embodiment, the step of generating target reply information corresponding to the user question description information based on the question and answer pair corresponding to the target data vector comprises: determining a model prompt word based on the user question description information, the question and answer pair corresponding to the target data vector, and an information output rule; and inputting the model prompt word into a pre-trained large language model to obtain the target reply information corresponding to the user question description information.

[0095] In an optional embodiment, after the step of obtaining the user question description information and the user input information, the method further comprises: determining whether the user question description information is valid information; if it is valid information, performing the step of inputting the user question description information and the user input information into the preset trained data processing model; and if it is not valid information, returning preset reply information to the user device sending the user question description information.

[0096] In an optional embodiment, after the steps of obtaining the user problem description information and the user input information, the method further comprises: inputting the user problem description information into the pre-trained label model to obtain a target label corresponding to the user problem description information; and wherein the target label is a label in the plurality of preset labels.

[0097] In an optional embodiment, after the steps of obtaining the user problem description information and the user input information, the method further comprises: identifying a first language corresponding to the user problem description information. After the step of generating the target reply information corresponding to the user problem description information based on the question and answer reply information corresponding to the target data vector, the method further comprises: identifying a second language corresponding to the target reply information corresponding to the user problem description information; determining whether the first language is the same as the second language; if yes, sending the target reply information to the user equipment that sends the user problem description information; and if no, translating the target reply information into the first language, and returning the target reply information in the first language to the user equipment that sends the user problem description information.

[0098] Further, Figure 4 The electronic device shown further includes a bus 102 and a communication interface 103, and the processor 101, the communication interface 103 and the memory 100 are connected through the bus 102.

[0099] The memory 100 can include a high-speed random access memory (RAM) and can also include a non-volatile memory such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 103 (which can be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. can be used. The bus 102 can be an ISA bus, a PCI bus or an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one bidirectional arrow is used in the figure to represent only one bus or one type of bus.

[0100] The processor 101 can be an integrated circuit chip with processing capability. In the implementation process, the steps of the above method can be completed by the integrated logic circuit or the instruction in the form of software in the processor 101. The processor 101 described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present disclosure can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiments of the present disclosure can be directly embodied as a hardware code processor for execution, or a combination of hardware and software modules in the code processor for execution. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium in the art. The storage medium is located in the memory 100, and the processor 101 reads the information in the memory 100, and combines the hardware to complete the steps of the method of the above embodiment.

[0101] The embodiments of the present disclosure also provide a computer readable storage medium, which stores computer executable instructions. When the computer executable instructions are called and executed by a processor, the computer executable instructions cause the processor to implement the above data processing method.

[0102] Specifically, the above data processing method comprises: obtaining user problem description information and user input information; wherein the user input information is used to indicate the information sending state and / or log information corresponding to the user problem description information; inputting the user problem description information and the user input information into a preset data processing model trained, to obtain an output vector; selecting a target data vector matched with the output vector from a preset question and answer library, and determining the question and answer information corresponding to the target data vector in the preset question and answer library; wherein the preset question and answer library comprises a plurality of question and answer pairs, and data vectors corresponding to the question and answer pairs; the question and answer pair comprises question description information and question description information corresponding question and answer information; and generating target answer information corresponding to the user problem description information based on the question and answer information corresponding to the target data vector.

[0103] In the data processing method, based on the user question description information and the user input information, a more accurate question and answer record can be recalled from the preset question and answer library, and question and answer information is generated based on the question and answer record, so that the user's intention can be accurately understood and a correct reply can be given, and the efficiency of customer service question reply can be improved.

[0104] In an optional embodiment, the step of selecting the target data vector matching the output vector from the preset question and answer library comprises: determining the similarity of the output vector and the data vector corresponding to each question and answer pair in the preset question and answer library; and selecting the target data vector matching the output vector from the preset question and answer library based on the similarity.

[0105] In an optional embodiment, the step of selecting the target data vector matching the output vector from the preset question and answer library based on the similarity comprises one of the following: determining the data vector with a similarity greater than a first preset similarity threshold in the preset question and answer library as the target data vector matching the output vector; and sorting the similarity of the output vector and the data vector corresponding to each question and answer pair in the preset question and answer library in descending order to obtain a sorting result, and determining the data vectors with a specified number of top sorting results as the target data vectors matching the output vector.

[0106] In an optional embodiment, the preset question and answer library is determined by: obtaining a plurality of historical reply records; wherein the historical reply record comprises a question and answer pair composed of question description information and question and answer information; for each historical reply record in the plurality of historical reply records, inputting the historical reply record into a data processing model to obtain a data vector corresponding to the historical reply record; determining the similarity between the data vectors corresponding to each two historical reply records in the plurality of historical reply records; dividing the historical reply records with a similarity less than a second preset threshold into the same clustering group to obtain a plurality of clustering groups; wherein the clustering group comprises at least one of the plurality of historical reply records and the data vectors corresponding to the plurality of historical reply records; and determining the plurality of clustering groups as the preset question and answer library.

[0107] In an optional embodiment, the method further comprises: grouping the user question description information and the target answer information into a target question and answer pair, and determining the output vector as a target data vector corresponding to the target question and answer pair; determining a target clustering group to which the target data vector belongs in the preset question and answer library based on the similarity of the target data vector and the data vector corresponding to each question and answer pair in the preset question and answer library; and storing the target question and answer pair and the target data vector in the target clustering group in association to update the preset question and answer library.

[0108] In an optional embodiment, the step of selecting, from the preset question and answer library, the target data vector matching the output vector based on the similarity, comprises: selecting, from each cluster group in the preset question and answer library, a first data vector with the highest similarity to the output vector; determining the first data vector in each cluster group as the target data vector matching the output vector; and wherein the target data vector contains a number of vectors matching the number of cluster groups contained in the preset question and answer library.

[0109] In an optional embodiment, the step of generating the target reply information corresponding to the user problem description information based on the question and reply information corresponding to the target data vector, comprises: determining a model prompt word based on the user problem description information, the question and reply information corresponding to the target data vector, and an information output rule; and inputting the model prompt word into a pre-trained large language model to obtain the target reply information corresponding to the user problem description information.

[0110] In an optional embodiment, after the step of obtaining the user problem description information and the user input information, the method further comprises: determining whether the user problem description information is valid information; if the user problem description information is valid information, performing the step of inputting the user problem description information and the user input information into the pre-trained data processing model; and if the user problem description information is not valid information, returning preset reply information to the user equipment sending the user problem description information.

[0111] In an optional embodiment, after the step of obtaining the user problem description information and the user input information, the method further comprises: inputting the user problem description information into a pre-trained label model to obtain a target label corresponding to the user problem description information; and wherein the target label is a label in a plurality of preset labels.

[0112] In an optional embodiment, after the step of obtaining the user problem description information and the user input information, the method further comprises: identifying a first language corresponding to the user problem description information. After the step of generating the target reply information corresponding to the user problem description information based on the question and reply information corresponding to the target data vector, the method further comprises: identifying a second language corresponding to the target reply information corresponding to the user problem description information; determining whether the first language is the same as the second language; if the first language is the same as the second language, returning the target reply information to the user equipment sending the user problem description information; and if the first language is not the same as the second language, translating the target reply information into the first language, and returning the target reply information in the first language to the user equipment sending the user problem description information.

[0113] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present disclosure can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a terminal device, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present disclosure. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0114] In the description of the present disclosure, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present disclosure and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present disclosure. In addition, the terms "first", "second", "third" are only for the purpose of description, and cannot be understood as indicating or implying relative importance.

[0115] Finally, it should be noted that: the above-described embodiments are only specific embodiments of the present disclosure, used to illustrate the technical solutions of the present disclosure, and not to limit them, the protection scope of the present disclosure is not limited thereto, although the present disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand: any person skilled in the art within the technical range disclosed by the present disclosure, they can still modify or easily think of changes to the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be covered within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. A data processing method, characterized in that, The method includes: The system acquires user problem description information and user input information; wherein the user input information is used to indicate the information sending status and / or log information corresponding to the user problem description information; the user problem description information is the problem text submitted by the user through the user device, and the user input information includes one or more of the following: user status information, user behavior information, and recorded log information; wherein the user status information is used to indicate the device status when the user sends the user problem description information, the user behavior information is used to indicate the behavior performed by the user on the user device when or before sending the user problem description information, and the log information is used to record error or vulnerability information; The user question description information and the user input information are input into a pre-trained data processing model to obtain an output vector; A target data vector matching the output vector is selected from a preset question-and-answer database, and the question-and-answer information corresponding to the target data vector in the preset question-and-answer database is determined; wherein, the preset question-and-answer database includes multiple question-and-answer pairs and data vectors corresponding to the question-and-answer pairs; the question-and-answer pairs include question description information and question-and-answer information corresponding to the question description information; Based on the question and answer information corresponding to the target data vector, target answer information corresponding to the user question description information is generated.

2. The method according to claim 1, characterized in that, The step of selecting a target data vector from a preset question-and-answer database that matches the output vector includes: Determine the similarity between the output vector and the data vector corresponding to each question-answer pair in the preset question-answer database; Based on the similarity, a target data vector that matches the output vector is selected from the preset question-and-answer database.

3. The method according to claim 2, characterized in that, The step of selecting a target data vector that matches the output vector from the preset question-and-answer database based on the similarity includes one of the following methods: Data vectors in the preset question-and-answer database with a similarity greater than a first preset similarity threshold are identified as target data vectors that match the output vector; and, The similarity between the output vector and the corresponding data vectors of the question answers in the preset question-answer database is sorted in descending order to obtain a sorting result. A specified number of data vectors at the top of the sorting result are determined as target data vectors that match the output vector.

4. The method according to claim 2, characterized in that, The preset question-and-answer database is determined in the following way: Retrieve multiple historical response records; wherein, the historical response records include: question-answer pairs consisting of question description information and question-answer information; For each of the plurality of historical reply records, the historical reply record is input into the data processing model to obtain the data vector corresponding to the historical reply record; Determine the similarity between the data vectors corresponding to every two of the plurality of historical reply records; The historical reply records with similarity less than a second preset threshold are divided into the same cluster group to obtain multiple cluster groups; wherein, each cluster group includes at least one of the multiple historical reply records and a data vector corresponding to each of the multiple historical reply records; The multiple cluster groups are identified as the preset question and answer database.

5. The method according to claim 4, characterized in that, The method further includes: The user question description information and the target answer information are combined to form a target question answer pair, and the output vector is determined as the target data vector corresponding to the target question answer pair; Based on the similarity between the target data vector and the data vector corresponding to each question-answer pair in the preset question-answer database, the target cluster group to which the target data vector belongs in the preset question-answer database is determined; The target question-answer pair and the target data vector are associated and stored in the target cluster group to update the preset question-answer database.

6. The method according to claim 4, characterized in that, The step of selecting a target data vector that matches the output vector from the preset question-answering database based on the similarity includes: From each cluster group in the preset question-and-answer database, select the first data vector with the highest similarity to the output vector; The first data vector in each cluster group is determined as the target data vector that matches the output vector; wherein the number of vectors contained in the target data vector matches the number of cluster groups contained in the preset question-and-answer database.

7. The method according to claim 1, characterized in that, The step of generating target response information corresponding to the user question description information based on the question response information corresponding to the target data vector includes: Based on the user question description information, the question answer information corresponding to the target data vector, and the information output rules, determine the model prompt words; The model prompts are input into a pre-trained large language model to obtain the target response information corresponding to the user's question description information.

8. The method according to claim 1, characterized in that, After the steps of obtaining user problem description information and user input information, the method further includes: Determine whether the user's problem description information is valid; If the information is valid, proceed with the step of inputting the user question description information and the user input information into the preset trained data processing model; If the information is not valid, a preset response message is returned to the user device that sent the user's problem description information.

9. The method according to claim 1, characterized in that, After the steps of obtaining user problem description information and user input information, the method further includes: The user problem description information is input into a pre-trained label model to obtain the target label corresponding to the user problem description information; wherein, the target label is a label from a plurality of preset labels.

10. The method according to claim 1, characterized in that, After the steps of obtaining user problem description information and user input information, the method further includes: Identify the first language corresponding to the user's problem description information; After the step of generating target response information corresponding to the user question description information based on the question response information corresponding to the target data vector, the method further includes: Identify the second language corresponding to the target response information corresponding to the user's question description information; Determine whether the first language is the same as the second language; if they are the same, send the target response information to the user device that sent the user problem description information. If they are different, the target response information is translated into the first language, and the target response information in the first language is returned to the user device that sent the user problem description information.

11. A data processing apparatus, characterized in that, The device includes: An information acquisition module is used to acquire user problem description information and user input information; wherein, the user input information is used to indicate the information sending status and / or log information corresponding to the user problem description information; the user problem description information is the problem text submitted by the user through the user device, and the user input information includes one or more of the following: user status information, user behavior information, and recorded log information; wherein, the user status information is used to indicate the device status when the user sends the user problem description information, the user behavior information is used to indicate the behavior performed by the user on the user device when or before sending the user problem description information, and the log information is used to record error or vulnerability information; The vector determination module is used to input the user question description information and the user input information into a preset trained data processing model to obtain an output vector; The response determination module is used to select a target data vector that matches the output vector from a preset question-and-answer database, and determine the question-and-answer information corresponding to the target data vector in the preset question-and-answer database; wherein, the preset question-and-answer database includes multiple question-and-answer pairs and data vectors corresponding to the question-and-answer pairs; the question-and-answer pairs include question description information and question-and-answer information corresponding to the question description information; The response generation module is used to generate target response information corresponding to the user question description information based on the question response information corresponding to the target data vector.

12. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the data processing method according to any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the data processing method according to any one of claims 1 to 10.

Images