A question and answer recommendation method, system, device, storage medium and program product

Abstract

The application discloses a kind of question and answer recommendation method, system, equipment, storage medium and program product, the method includes: real-time acquisition first information, first information contains first inquiry information and multiple second inquiry information, the inquiry time corresponding to second inquiry information is earlier than the inquiry time corresponding to first inquiry information;First information is carried out semantic analysis, determine first demand feature, first demand feature is used to describe first information in from at least one of the following characterization dimensions: first characterization dimension, second characterization dimension and third characterization dimension;First demand feature is searched in first vector knowledge base based on, determine multiple response result data for first inquiry information;Wherein, the similarity between response result data and first demand feature is greater than similarity threshold, and first vector knowledge base is formed by the response vector data corresponding to the response data of multiple different modalities.The present scheme can realize automatic and fast accurate question and answer recommendation.

Description

Technical Field

[0001] This invention relates to the field of computer processing technology, and in particular to a question-and-answer recommendation method, system, device, storage medium, and program product. Background Technology

[0002] In some WeChat Work customer service scenarios, when responding to user questions, WeChat Work customer service staff need to manually extract and input keywords from the user's question in order to retrieve FAQ (Frequently Asked Questions) content from the FAQ knowledge base and then provide a response to the user. However, this approach is extremely limited by the accuracy of the keywords manually extracted by the WeChat Work customer service staff, and heavily relies on the staff's experience and familiarity with the FAQ knowledge base. This can easily lead to problems such as low search result hit rate, low overall search efficiency, and inconsistent user experience. Summary of the Invention

[0003] This invention provides a question-and-answer recommendation method, system, device, storage medium, and program product, which can achieve automatic, fast, and accurate question-and-answer recommendation.

[0004] In a first aspect, the present invention provides a question-and-answer recommendation method, comprising: The first information is acquired in real time, which includes a first query and multiple second query messages, wherein the query time corresponding to the second query message is earlier than the query time corresponding to the first query message. Semantic analysis is performed on the first information to determine the first demand feature. The first demand feature is used to characterize the first information from at least one of the following characterization dimensions: a first characterization dimension, a second characterization dimension, and a third characterization dimension. The first characterization dimension is used to indicate the query demand content corresponding to the first information, the second characterization dimension is used to indicate the dialogue scenario to which the first information belongs, and the third characterization dimension is used to indicate the entity objects involved in the first information. Based on the first demand feature, a search is performed in the first vector knowledge base to determine multiple response result data for the first query information; wherein, the similarity between the response result data and the first demand feature is greater than a similarity threshold, and the first vector knowledge base is composed of response vector data corresponding to response data of multiple different modalities.

[0005] Secondly, the present invention also provides a question-answering recommendation system, the system comprising a question-answering recommendation device, a computing power service device, and a first vector knowledge base; wherein: The question-and-answer recommendation device is used to acquire first information in real time. The first information includes a first question and multiple second questions. The question time corresponding to the second question is earlier than the question time corresponding to the first question. The question-and-answer recommendation device is further configured to send the first information to the computing power service device, so that the computing power service device performs the following operations: performing semantic analysis on the first information to determine a first demand feature; and retrieving multiple response result data for the first query information based on the first demand feature in a first vector knowledge base; wherein, the first demand feature is used to characterize the first information from at least one of the following characterization dimensions: a first characterization dimension, a second characterization dimension, and a third characterization dimension; the first characterization dimension is used to indicate the query demand content corresponding to the first information, the second characterization dimension is used to indicate the dialogue scenario to which the first information belongs, and the third characterization dimension is used to indicate the entity objects involved in the first information; the similarity between the response result data and the first demand feature is greater than a similarity threshold, and the first vector knowledge base is composed of response vector data corresponding to response data of multiple different modalities.

[0006] The question-and-answer recommendation device is also used to receive multiple response result data sent by the computing power service device.

[0007] Thirdly, this embodiment of the invention also provides a question-and-answer recommendation device, the question-and-answer recommendation device comprising: One or more processors; Storage device 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 implement the question-and-answer recommendation method as provided in any embodiment of the present invention.

[0008] Fourthly, embodiments of the present invention also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the question-and-answer recommendation method provided in any embodiment of the present invention.

[0009] Fifthly, this invention also provides a computer program product, including a computer program that, when executed by a processor, implements the question-and-answer recommendation method provided in any embodiment of this invention.

[0010] The technical solution of this invention acquires first information in real time, which includes a first query and multiple second queries, wherein the query time of the second queries is earlier than that of the first queries, thereby timely capturing the specific query content of the user's dialogue flow context. Then, by performing semantic analysis on the first information, a first demand feature is determined. This first demand feature is used to characterize the first information from at least one of the following dimensions: a first characterization dimension, a second characterization dimension, and a third characterization dimension. The first characterization dimension indicates the query demand content corresponding to the first information, the second characterization dimension indicates the dialogue scenario to which the first information belongs, and the third characterization dimension indicates the entity objects involved in the first information, thus realizing the parsing of the user's demand features and core intent from the contextual information content. Next, based on the first demand feature, a search is performed in a first vector knowledge base to determine multiple response result data for the first query. The similarity between the response result data and the first demand feature is greater than a similarity threshold. The first vector knowledge base is composed of response vector data corresponding to response data of various different modalities. This solution allows for core intent analysis based on multiple real-time query information, thereby identifying user needs and characteristics. Furthermore, based on these needs and characteristics, a precise search is performed in the first vector knowledge base to determine multiple response results. This enables automatic, fast, and accurate question-and-answer recommendations, significantly improving the response speed of WeChat customer service and the accuracy of FAQ responses.

[0011] The above description of the invention is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0013] Figure 1 This is a flowchart illustrating a question-and-answer recommendation method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating another question-and-answer recommendation method provided in an embodiment of the present invention; Figure 3 A schematic diagram of the logic flow of a question-answering recommendation method provided in an embodiment of the present invention; Figure 4This is a schematic diagram of the structure of a question-answering recommendation system provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a question-answering recommendation device that implements a question-answering recommendation method according to an embodiment of the present invention. Detailed Implementation

[0014] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0015] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0016] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0017] It should be noted that the concepts of "first" and "second" mentioned in this invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0018] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0019] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0020] Figure 1 This is a flowchart illustrating a question-and-answer recommendation method provided in an embodiment of the present invention. This embodiment is applicable to situations in WeChat customer service scenarios where user inquiries are addressed by determining or recommending response content. This method can be executed by a question-and-answer recommendation device, which can be a mobile terminal, a PC, or a server, etc. Figure 1 As shown, the question-and-answer recommendation method of this invention may include the following process: S110. Obtain first information in real time. The first information includes first query information and multiple second query information. The query time corresponding to the second query information is earlier than the query time corresponding to the first query information.

[0021] The first information can be understood as a collection of information consisting of multiple inquiries. This first information includes a first inquiry and multiple second inquiries, which refer to information obtained through inquiries in order to obtain corresponding answers. Both the first and second inquiries can be in text format, both in voice format, or a combination of both. For example, in online consultation services such as medical health and preventative care provided by the WeChat Work platform, the first and second inquiries can be initiated by the user. For instance, a user could inquire with WeChat Work customer service, specifically asking "What should I pay attention to after surgery for pleural effusion?" Such text content would then constitute an inquiry. Of course, users can also inquire in this scenario by sending voice messages.

[0022] Accordingly, the query time is the moment the query message is sent, and each query message corresponds to a query time. It's understandable that in such a query scenario, a user might not only make one query, but continuously launch multiple queries over time, resulting in multiple query messages. In this embodiment, the query time corresponding to the second query message is earlier than the query time corresponding to the first query message. That is to say, the first query message is essentially the one with the latest query time, and is essentially the query message that currently needs a response / reply, while multiple second query messages are essentially historical query messages that preceded the first query message.

[0023] Specifically, this can be achieved by continuously monitoring the conversation flow within the customer service representative's interface to obtain the user's initial inquiry and multiple subsequent inquiries, thus acquiring the initial information in real time. Of course, the time of each inquiry will also be simultaneously obtained.

[0024] S120. Perform semantic analysis on the first information to determine the first demand feature. The first demand feature is used to characterize the first information from at least one of the following characterization dimensions: first characterization dimension, second characterization dimension, and third characterization dimension. The first characterization dimension is used to indicate the query demand content corresponding to the first information, the second characterization dimension is used to indicate the dialogue scenario to which the first information belongs, and the third characterization dimension is used to indicate the entity objects involved in the first information.

[0025] In this context, the characterization dimension can be understood as the dimension used to describe or characterize the first information. In this embodiment, the characterization dimension includes at least one of the following: a first characterization dimension, a second characterization dimension, and a third characterization dimension. The first characterization dimension indicates the content of the inquiry request corresponding to the first information; for example, such an inquiry request could be a consultation process, policy inquiry, report interpretation, etc. The second characterization dimension indicates the dialogue scenario to which the first information belongs; for example, such a dialogue scenario could be charitable drug donation, pre-consultation, triage, follow-up visit, etc. The third characterization dimension indicates the entity objects involved in the first information; for example, the involved entity objects could be drugs, diseases, departments, adverse reactions, etc. It should be noted that the various specific details involved in the first, second, and third characterization dimensions can be further categorized based on actual needs. The above are just some examples; more examples will not be detailed here. Of course, in addition to the first, second, and third characterization dimensions, more characterization dimensions can be set based on the needs of the actual scenario and the characterization objectives.

[0026] Correspondingly, the first requirement feature is used to characterize the first information from at least one of the first, second, and third characterization dimensions. This first requirement feature essentially reflects the core intent corresponding to the first information.

[0027] Specifically, the acquired initial information can be semantically analyzed to determine the initial demand features. For example, a natural language processing model can be used to perform semantic analysis on the initial information to obtain the initial demand features. The determined initial demand features can be in the form of structured data labels or in the form of high-dimensional vectors.

[0028] S130. Based on the first demand feature, a search is performed in the first vector knowledge base to determine multiple response result data for the first query information; wherein, the similarity between the response result data and the first demand feature is greater than the similarity threshold, and the first vector knowledge base is composed of response vector data corresponding to response data of multiple different modalities.

[0029] In this context, response data refers to data that can be used as answer content. Specifically, response data can be FAQ (Frequently Asked Questions) data, which can be standardized and adapted to the specific scenario of this question-and-answer recommendation method. For example, in a medical or health consultation scenario, this response data could be considered a medical-related FAQ. Correspondingly, modality refers to the type of data; for example, text data is one modality, image data is another, and video data is yet another. Response vector data refers to the data obtained after vectorizing the response data.

[0030] In this embodiment, the first vector knowledge base is composed of response vector data corresponding to response data of various different modalities. These response data can cover multiple different modalities, such as text-based FAQ data, video-based FAQ data, or even image-based FAQ data. The first vector knowledge base is composed of the response vector data corresponding to each of these response data.

[0031] Specifically, after identifying the primary demand characteristic, a search can be performed in the first vector knowledge base based on this characteristic to retrieve data with a similarity greater than a similarity threshold. This allows for the determination of multiple response results for the primary query. The similarity threshold can be adjusted based on actual needs. These multiple response results are then displayed on the interface, enabling WeChat customer service personnel to respond to the user's current inquiry. Customer service personnel can choose one or more responses from the displayed results to reply to.

[0032] As an optional but non-limiting implementation, the first vector knowledge base is constructed as follows: Multiple response data points are acquired, with at least some of the response data points belonging to different modalities; the multiple response data points are embedded using an embedding processing model to obtain multiple response vector data points; a response vector index is constructed based on the multiple response vector data points; and the first vector knowledge base is constructed based on the response vector index and the multiple response vector data points. This optional approach allows for the vectorization of multiple response data points of different modalities to construct a comprehensive vector knowledge base, facilitating the accurate retrieval of various types of response data in question-and-answer scenarios in subsequent solutions.

[0033] The embedding processing model can be used to transform unstructured data into high-dimensional vector data. In this embodiment, the embedding processing model can be a pre-trained model capable of embedding response data from different modalities and mapping the processed response data from different modalities to the same semantic space. That is, this embedding processing model can not only transform text data into high-dimensional vector data, but also image data into high-dimensional vector data, etc., and after embedding, the response vector data obtained from different modalities are represented in the same semantic space.

[0034] The response vector index can be understood as a data structure that facilitates similarity searches and nearest neighbor retrieval in the first vector knowledge base, thereby improving retrieval efficiency in the first vector knowledge base.

[0035] Specifically, multiple response data points can be identified from various data sources based on the needs of the actual scenario. At least some of these response data points should be of different modalities, thus obtaining multiple response data points. Then, by embedding these multiple response data points using an embedding processing model, response vector data corresponding to each response data point can be obtained, resulting in multiple response vector data points. Next, a response vector index can be constructed based on these multiple response vector data points. For example, a graph-structured response vector index can be constructed using the HNSW (Hierarchical Navigable Small World) algorithm based on the relationships between the multiple response vector data points, or a tree-structured response vector index can be constructed using the KD-Tree algorithm; alternatively, a clustering-based response vector index can be constructed using the IVF (Inverted File Index) algorithm based on the clustering relationships between the multiple response vector data points. Subsequently, a first vector knowledge base can be constructed based on the response vector index and the multiple response vector data points. Of course, this first vector knowledge base can be continuously updated and optimized during subsequent use.

[0036] The technical solution of this invention acquires first information in real time, which includes a first query and multiple second queries, wherein the query time of the second queries is earlier than that of the first queries, thereby timely capturing the specific query content of the user's dialogue flow context. Then, by performing semantic analysis on the first information, a first demand feature is determined. This first demand feature is used to characterize the first information from at least one of the following dimensions: a first characterization dimension, a second characterization dimension, and a third characterization dimension. The first characterization dimension indicates the query demand content corresponding to the first information, the second characterization dimension indicates the dialogue scenario to which the first information belongs, and the third characterization dimension indicates the entity objects involved in the first information, thus realizing the parsing of the user's demand features and core intent from the contextual information content. Next, based on the first demand feature, a search is performed in a first vector knowledge base to determine multiple response result data for the first query. The similarity between the response result data and the first demand feature is greater than a similarity threshold. The first vector knowledge base is composed of response vector data corresponding to response data of various different modalities. This solution allows for core intent analysis based on multiple real-time query information, thereby identifying user needs and characteristics. Furthermore, based on these needs and characteristics, a precise search is performed in the first vector knowledge base to determine multiple response results. This enables automatic, fast, and accurate question-and-answer recommendations, significantly improving the response speed of WeChat customer service and the accuracy of FAQ responses.

[0037] Figure 2 This is a flowchart illustrating another question-answering recommendation method provided by an embodiment of the present invention. The technical solution of this embodiment further optimizes the process of semantic analysis of the first information to determine the first demand feature in the aforementioned embodiments, based on the technical solutions of the above embodiments. This embodiment can be combined with various optional solutions in one or more of the above embodiments. Figure 2 As shown, the question-and-answer recommendation method of this invention may include the following process: S210. Obtain first information in real time. The first information includes first query information and multiple second query information. The query time corresponding to the second query information is earlier than the query time corresponding to the first query information.

[0038] S220. Based on the characterization dimension, a natural language processing model is used to extract information from the first query information and multiple second query information to obtain a set of field information; wherein, the set of field information includes at least one of the following fields: requirement target field, scenario theme field, and entity name field.

[0039] Natural language processing (NLP) models can be used to perform semantic analysis and extract key information from natural language text. In this embodiment, the NLP model can be a model trained using characterization dimensions, which, after training, can extract key information belonging to the characterization dimensions. For example, the NLP model can be a trained BERT (Bidirectional Encoder Representations from Transformers) model, etc.

[0040] Specifically, based on the characterization dimensions, a natural language processing model can be used to extract information from the first query and multiple second queries, thereby obtaining the fields corresponding to each query. The sum of all fields yields a set of field information. This set of field information includes at least one of the following fields: a demand target field, a scenario theme field, and an entity name field. The demand target field corresponds to the first characterization dimension and describes the specific query demand content of the first information within that dimension. The scenario theme field corresponds to the second characterization dimension and describes the specific dialogue scenario to which the first information belongs within that dimension. The entity name field corresponds to the third characterization dimension and describes the specific entity object involved in the first information within that dimension. For example, after semantic analysis and information extraction, the resulting set of field information could include the following fields: "pleural effusion," "scientific understanding," "late-stage management," and "coping strategies."

[0041] Optionally, before using a natural language processing model to extract information from the first query information and multiple second query information respectively, the method further includes: performing preprocessing operations on the first query information and multiple second query information respectively. These preprocessing operations include, but are not limited to: pause word removal processing, special character filtering processing, spelling and grammar error correction processing, and colloquial expression conversion processing.

[0042] It should be noted that when the query information is in the form of voice, it can be converted into text information first, and then a natural language processing model can be used for semantic analysis and field extraction.

[0043] S230. The embedding processing model is used to perform text embedding processing on each field in the field information set to obtain multiple text vectors.

[0044] The embedding processing model can be used to transform unstructured data into high-dimensional vector data. It should be noted that in this embodiment, the embedding processing model is the same one used in the first vector knowledge base construction process. By using the same embedding processing model to process the fields corresponding to the first information, the processed vector data can be kept in the same semantic space as the existing response vector data in the first vector knowledge base, facilitating data retrieval in subsequent schemes.

[0045] Specifically, an embedding processing model can be used to perform text embedding processing on each field in the field information set to obtain the text vector corresponding to each field, thereby obtaining multiple text vectors.

[0046] S240. Concatenate multiple text vectors to determine the first required feature.

[0047] Specifically, after obtaining the text vectors corresponding to each field, multiple text vectors can be concatenated to determine the first required feature.

[0048] S250. Based on the first demand feature, a search is performed in the first vector knowledge base to determine multiple response result data for the first query information; wherein, the similarity between the response result data and the first demand feature is greater than the similarity threshold, and the first vector knowledge base is composed of response vector data corresponding to response data of multiple different modalities.

[0049] As an optional but non-limiting implementation, a search is performed in a first vector knowledge base based on the first requirement feature to determine multiple response result data for the first query information. This includes: determining a set of candidate response vectors from the first vector knowledge base based on the first requirement feature and the response vector index; wherein the set of candidate response vectors contains at least one candidate response vector data; calculating the cosine similarity between each candidate response vector data and the first requirement feature, and determining multiple target response vector data based on multiple cosine similarities and similarity thresholds; and determining multiple response result data based on the response data corresponding to each target response vector data. Using this optional scheme, multiple response result data that meet the requirements can be retrieved and determined from the first vector knowledge base based on the first requirement feature.

[0050] The candidate response vector set can be understood as a set of vectors initially determined by the response vector index, and the candidate response vector set contains at least one candidate response vector data.

[0051] Specifically, a search can be performed in a first vector knowledge base based on the first requirement feature and the response vector index to determine a set of candidate response vectors. For example, the Approximate Nearest Neighbor Search (ANNS) algorithm can be used to determine the set of candidate response vectors in the first vector knowledge base. Then, the cosine similarity between each candidate response vector data in this set and the first requirement feature can be calculated. Based on the obtained multiple cosine similarities and similarity thresholds, multiple target response vector data can be determined from these candidate response vector data, wherein the cosine similarity between each target response vector data and the first requirement feature is greater than the set similarity threshold. Finally, multiple response result data can be determined based on the response data corresponding to each target response vector data.

[0052] As an optional but non-limiting implementation, after determining multiple response results for the first query, the method further includes: determining the weight coefficient of each response result under a preset evaluation dimension; the preset evaluation dimension includes at least one of the following: data recommendation dimension, usage frequency dimension, and project affiliation dimension; determining the score of each response result based on the weight coefficient; and determining a preset number of response results to be displayed from the multiple response results based on the score, and recommending the response results to be displayed. Using this optional solution, each response result can be evaluated based on different preset evaluation dimensions, thereby determining the response results to be displayed that meet the recommendation requirements.

[0053] The preset evaluation dimensions can be understood as dimensions used to evaluate and analyze the response result data. In this embodiment, the preset evaluation dimensions include at least one of the following: data recommendation dimension, usage frequency dimension, and project affiliation dimension. The data recommendation dimension is used to indicate the degree of recommendation for the response result data. The usage frequency dimension is used to indicate the frequency of acquisition or usage of the response result data within a preset time period (e.g., the most recent week, the most recent month, etc.). The project affiliation dimension is used to indicate the project to which the response result data belongs; for example, a response result data a1 belongs to medical project A, while a response result data b1 belongs to medical project B.

[0054] Understandably, different specific evaluation results under the same preset evaluation dimension will correspond to different weight coefficients. For example, under the "project affiliation dimension," if the specific evaluation result corresponding to response result data a1 is determined to be "belongs to medical project A," and the specific evaluation result corresponding to response result data b1 is determined to be "belongs to medical project B," and assuming that medical project A is more important than medical project B and belongs to projects that need to be promoted in the near future, then a relatively higher weight coefficient (e.g., 0.6) will be assigned to response result data a1 under the "project affiliation dimension," while a relatively lower weight coefficient (e.g., 0.3) will be assigned to response result data b1 under the "project affiliation dimension." The same principle applies to other preset evaluation dimensions, which will not be elaborated here.

[0055] Specifically, in this embodiment, different weight coefficients can be pre-set for the specific evaluation results under each preset evaluation dimension. Then, based on the specific evaluation results of each response result data under each preset evaluation dimension, the weight coefficient of each response result data under each preset evaluation dimension can be determined. For example, for a certain response result data c1, its coefficient can be determined to be 0.5 in the "data recommendation dimension," 0.7 in the "usage frequency dimension," and 0.4 in the "project affiliation dimension," etc. Then, based on each weight coefficient, the score result of each response result data can be determined. For example, it can be determined by weighted summation of each weight coefficient. Subsequently, based on each score result, a preset number of response result data to be displayed can be determined from multiple response result data, and these data can be recommended for display. This preset number can be set based on actual needs. For example, the score results can be sorted from high to low, and the data with the top five scores (i.e., TOP-5) can be selected as the response result data to be displayed.

[0056] As an optional but non-limiting implementation, the proposed display of the response result data includes: for a single response result data to be displayed, determining multiple display element results corresponding to the response result data based on preset element requirements; wherein, the preset element requirements are used to indicate the display elements that need to be displayed when displaying the response result data; based on the multiple display element results and the response result data to be displayed, generating display data in a preset display format, and displaying the display data at a preset position on the interface. Using this optional solution, the response result data to be displayed can be processed according to the specific needs of the display, thereby displaying the resulting display data at a preset position on the interface.

[0057] The preset element requirements specify the elements to be displayed when presenting the response results data. These elements could include: title, brief summary, content type, etc. Furthermore, the preset element requirements can indicate not only which elements to display but also the arrangement of these elements, allowing for customization based on actual needs.

[0058] In this embodiment, the preset display format refers to displaying each piece of data in the form of cards. The preset location can be a certain side position of the workbench interface.

[0059] Specifically, after obtaining multiple response results to be displayed, for each individual response result, multiple display element results can be determined based on preset element requirements. For example, after word segmentation, the display element result for a certain response result m under the display element "Title" is determined to be "Facing pleural effusion, there is no need to panic excessively; scientific understanding and response help in the management of late-stage lung cancer," and the display element result under the display element "Content Type" is "Article," etc. Then, based on the multiple display element results and the response results to be displayed, rendering processing can be performed to generate display data in a preset display format, and the display data can be displayed at a preset location on the interface. For example, each response result can be pushed to the customer service workbench of the "Enterprise WeChat Dialogue Interface" in the form of structured cards, with each card containing related elements such as title, brief summary, and content type. Of course, a "one-click send" button can also be configured in this structured card, making it convenient for customer service personnel to select one or more from the multiple structured cards displayed to reply to the user's current inquiry.

[0060] The technical solution of this invention acquires first information in real time, which includes a first query and multiple second queries, wherein the query time of the second queries is earlier than that of the first queries, thereby timely capturing the specific query content of the user's dialogue flow context. Then, based on the characterization dimension, a natural language processing model is used to extract information from the first query and multiple second queries to obtain a set of field information. This set of field information includes at least one of the following fields: a demand target field, a scenario theme field, and an entity name field. An embedding processing model is used to perform text embedding processing on each field in the set of field information to obtain multiple text vectors. The multiple text vectors are concatenated to determine the first demand feature. This achieves the parsing of the user's demand feature and core intent from the context information content. Next, based on the first demand feature, a search is performed in a first vector knowledge base to determine multiple response result data for the first query. The similarity between the response result data and the first demand feature is greater than a similarity threshold. The first vector knowledge base is composed of response vector data corresponding to response data of various different modalities. This solution allows for core intent analysis based on multiple real-time query information, thereby identifying user needs and characteristics. Furthermore, based on these needs and characteristics, a precise search is performed in the first vector knowledge base to determine multiple response results. This enables automatic, fast, and accurate question-and-answer recommendations, significantly improving the response speed of WeChat customer service and the accuracy of FAQ responses.

[0061] Figure 3 This is a schematic diagram of the logical flow of a question-answering recommendation method provided in an embodiment of the present invention. Solutions not described in detail in this embodiment can be found in the above embodiments. Figure 3 As shown, the question-and-answer recommendation method is as follows: Step 1: The backend management terminal of the question-and-answer recommendation device platform will upload relevant FAQ data and connect to the computing power service device through a pre-configured interface, so that the computing power service device can embed the FAQ data and complete the construction of the vector knowledge base.

[0062] Step 2: The backend management terminal of the question-and-answer recommendation device will listen to the dialogue flow between the "user terminal" and the "customer terminal" in real time to obtain multiple query information from users in real time, and send the obtained multiple query information to the computing power service device through a preset interface.

[0063] Step 3: After receiving these queries, the computing power service device uses a natural language processing model to process them and obtain the user's needs / core intent. The determined needs can be in vector form. Next, the computing power service device can search in the established vector knowledge base to find the semantically most similar FAQ content vectors, thereby determining multiple FAQ recommendation data, and then sending these multiple FAQ recommendation data to the question-answering recommendation device platform.

[0064] Step 4: After receiving multiple FAQ recommendation data, the Q&A recommendation device platform will display them on the client-side interface of the Q&A recommendation device platform. For example, they can be displayed on the right side of the interface in the form of a structured card. The structured card displays elements such as title, brief summary, and content type, and also has a "one-click send" button.

[0065] Step 5: Customer service staff can then select one or more cards from the multiple cards displayed in the push notification to respond to the user's current inquiry.

[0066] This solution allows for core intent analysis based on multiple real-time query information, thereby identifying user needs and characteristics. Furthermore, based on these needs and characteristics, a precise search is performed in the first vector knowledge base to determine multiple response results. This enables automatic, fast, and accurate question-and-answer recommendations, significantly improving the response speed of WeChat customer service and the accuracy of FAQ responses.

[0067] Figure 4 This is a schematic diagram of a question-and-answer recommendation system provided in an embodiment of the present invention. This embodiment is applicable to situations in WeChat customer service scenarios where the system determines or recommends responses to user inquiries. Figure 4 As shown, the question-answering recommendation system of this embodiment may include a question-answering recommendation device 410, a computing power service device 420, and a first vector knowledge base 430. Wherein: The question-and-answer recommendation device 410 is used to acquire first information in real time. The first information includes first question information and multiple second question information. The question time corresponding to the second question information is earlier than the question time corresponding to the first question information. The question-and-answer recommendation device 410 is further configured to send the first information to the computing power service device 420, so that the computing power service device 420 performs the following operations: performing semantic analysis on the first information to determine a first demand feature; and, based on the first demand feature, performing a search in the first vector knowledge base 430 to determine multiple response result data for the first query information; wherein, the first demand feature is used to characterize the first information from at least one of the following characterization dimensions: a first characterization dimension, a second characterization dimension, and a third characterization dimension; the first characterization dimension is used to indicate the query demand content corresponding to the first information, the second characterization dimension is used to indicate the dialogue scenario to which the first information belongs, and the third characterization dimension is used to indicate the entity objects involved in the first information; the similarity between the response result data and the first demand feature is greater than a similarity threshold, and the first vector knowledge base 430 is composed of response vector data corresponding to response data of multiple different modalities.

[0068] The question-and-answer recommendation device 410 is also used to receive multiple response result data sent by the computing power service device 420.

[0069] The technical solution of this invention involves a question-and-answer recommendation device 410 acquiring first information in real time. This first information includes a first query and multiple second queries, where the query time of the second queries is earlier than that of the first queries, thus timely capturing the specific query content within the user's dialogue flow. The first information is then sent to a computing power service device 420, causing the computing power service device 420 to perform the following operations: semantic analysis of the first information to determine a first demand feature. This first demand feature is used to characterize the first information from at least one of the following dimensions: a first characterization dimension, a second characterization dimension, and a third characterization dimension. The first characterization dimension is used to indicate the first... The information corresponds to the query requirement content. The second characterization dimension is used to indicate the dialogue scenario to which the first information belongs, and the third characterization dimension is used to indicate the entity objects involved in the first information. This enables the parsing of the user's requirement characteristics and core intent from the contextual information content. Furthermore, the computing power service device 420 searches the first vector knowledge base 430 based on the first requirement characteristics to determine multiple response result data for the first query information. The similarity between the response result data and the first requirement characteristics is greater than a similarity threshold. The first vector knowledge base 430 is composed of response vector data corresponding to response data of various different modalities. The multiple response result data are then sent to the computing power service device 420. Using this solution, core intent analysis can be performed based on multiple query information acquired in real time to determine the user's requirement characteristics. Further, based on these requirement characteristics, a precise search can be performed in the first vector knowledge base to determine multiple response result data. This enables automatic, fast, and accurate question-and-answer recommendations, thereby significantly improving the response speed of enterprise WeChat customer service and the accuracy of FAQ data answers.

[0070] As an optional but non-limiting implementation, the computing power service device 420 includes a field information set determination module, an embedding processing module, and a first requirement feature determination module. Wherein: The field information set determination module is used to extract information from the first query information and multiple second query information based on the characterization dimension using a natural language processing model to obtain a field information set; wherein, the field information set includes at least one of the following fields: requirement target field, scenario theme field, and entity name field; The embedding processing module is used to perform text embedding processing on each field in the field information set using an embedding processing model to obtain multiple text vectors. The first requirement feature determination module is used to concatenate multiple text vectors to determine the first requirement feature.

[0071] As an optional but non-limiting implementation, the first vector knowledge base 430 is constructed in the following manner: acquiring multiple response data, at least some of which are data of different modalities; performing embedding processing on the multiple response data respectively through the embedding processing model to obtain multiple response vector data; constructing a response vector index based on the multiple response vector data; and constructing the first vector knowledge base based on the response vector index and the multiple response vector data.

[0072] As an optional but non-limiting implementation, the computing power service device 420 further includes: a candidate set determination module, a target response vector data determination module, and a response result data determination module. Wherein: The candidate set determination module is used to determine a candidate response vector set from the first vector knowledge base based on the first demand feature and the response vector index; wherein the candidate response vector set contains at least one candidate response vector data. The target response vector data determination module is used to calculate the cosine similarity between each candidate response vector data and the first requirement feature, and determine multiple target response vector data based on multiple cosine similarities and the similarity threshold. The response result data determination module is used to determine multiple response result data based on the response data corresponding to each target response vector data.

[0073] As an optional but non-limiting implementation, the question-answering recommendation device 410 includes a weight coefficient determination module, a scoring result determination module, and a recommendation display module. Wherein: The weight coefficient determination module is used to determine the weight coefficient of each response result data under a preset evaluation dimension; the preset evaluation dimension includes at least one of the following: data recommendation dimension, usage frequency dimension, and project affiliation dimension; The scoring result determination module is used to determine the scoring result of each of the response result data based on the weight coefficients; The recommendation display module is used to determine a preset number of response result data to be displayed from multiple response result data based on the scoring results, and to recommend and display the response result data to be displayed.

[0074] As an optional but non-limiting implementation, the recommendation display module includes a display element result determination unit and a display data generation unit. Wherein: The display element result determination unit is used to determine multiple display element results corresponding to a single response result data to be displayed, based on preset element requirements; wherein, the preset element requirements are used to indicate the display elements that need to be displayed when displaying the response result data to be displayed; The display data generation unit is used to generate display data in a preset display format based on the results of multiple display elements and the response results to be displayed, and to display the display data at a preset position on the interface.

[0075] The question-and-answer recommendation system provided in this embodiment of the invention can be used to execute question-and-answer recommendation methods, and has the corresponding functional modules and beneficial effects for executing question-and-answer recommendation methods.

[0076] It is worth noting that the various units and modules included in the above system are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of the present invention.

[0077] Figure 5 This is a schematic diagram of a question-answering recommendation device that implements a question-answering recommendation method according to an embodiment of the present invention. The following refers to... Figure 5 The diagram illustrates a structural schematic suitable for implementing the question-and-answer recommendation device 510 of the embodiments of the present invention. The question-and-answer recommendation device in the embodiments of the present invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The question-and-answer recommendation device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0078] like Figure 5 As shown, the question-and-answer recommendation device 510 includes at least one processor 511 and a memory, such as a read-only memory (ROM) 512 and a random access memory (RAM) 513, communicatively connected to the at least one processor 511. The memory stores computer programs executable by the at least one processor. The processor 511 can perform various appropriate actions and processes based on the computer program stored in the ROM 512 or loaded from storage unit 518 into the RAM 513. The RAM 513 may also store various programs and data required for the operation of the question-and-answer recommendation device 510. The processor 511, ROM 512, and RAM 513 are interconnected via a bus 514. An input / output (I / O) interface 515 is also connected to the bus 514.

[0079] Multiple components in the question-and-answer recommendation device 510 are connected to an input / output (I / O) interface 515, including: an input unit 516, such as a keyboard, mouse, etc.; an output unit 517, such as various types of displays, speakers, etc.; a storage unit 518, such as a disk, optical disk, etc.; and a communication unit 519, such as a network card, modem, wireless transceiver, etc. The communication unit 519 allows the question-and-answer recommendation device 510 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0080] Processor 511 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 511 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 511 executes the question-answering recommendation method provided in any embodiment of the present invention.

[0081] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the question-and-answer recommendation method shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 519, or installed from storage unit 518, or installed from read-only memory (ROM) 512. When the computer program is executed by processor 511, it performs the functions defined in the question-and-answer recommendation method of the embodiments of the present invention.

[0082] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0083] The question-and-answer recommendation device provided in this embodiment of the invention and the question-and-answer recommendation method provided in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0084] This invention provides a computer storage medium storing a computer program that, when executed by a processor, implements the question-and-answer recommendation method provided in the above embodiments.

[0085] It should be noted that the computer-readable medium described above in this invention can be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can 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. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0086] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0087] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0088] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0089] The units described in the embodiments of the present invention can be implemented in software or in hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0090] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0091] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0092] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0093] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in sequential order. In certain circumstances, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the invention. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0094] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A question-and-answer recommendation method, characterized in that, The method includes: The first information is acquired in real time, which includes a first query and multiple second query messages, wherein the query time corresponding to the second query message is earlier than the query time corresponding to the first query message. Semantic analysis is performed on the first information to determine the first demand feature. The first demand feature is used to characterize the first information from at least one of the following characterization dimensions: a first characterization dimension, a second characterization dimension, and a third characterization dimension. The first characterization dimension is used to indicate the query demand content corresponding to the first information, the second characterization dimension is used to indicate the dialogue scenario to which the first information belongs, and the third characterization dimension is used to indicate the entity objects involved in the first information. Based on the first demand feature, a search is performed in the first vector knowledge base to determine multiple response result data for the first query information; wherein, the similarity between the response result data and the first demand feature is greater than a similarity threshold, and the first vector knowledge base is composed of response vector data corresponding to response data of multiple different modalities.

2. The method according to claim 1, characterized in that, The step of performing semantic analysis on the first information to determine the first demand feature includes: Based on the aforementioned characterization dimension, a natural language processing model is used to extract information from the first query information and multiple second query information to obtain a set of field information; wherein, the set of field information includes at least one of the following fields: a requirement target field, a scenario theme field, and an entity name field; An embedding processing model is used to perform text embedding processing on each field in the field information set to obtain multiple text vectors; The first requirement feature is determined by concatenating multiple text vectors.

3. The method according to claim 2, characterized in that, The first vector knowledge base is constructed in the following manner: Acquire multiple response data, wherein at least some of the response data are data of different modalities; The embedding processing model is used to embed multiple response data into each response data to obtain multiple response vector data. Based on the multiple response vector data, a response vector index is constructed; Based on the response vector index and multiple response vector data, the first vector knowledge base is constructed.

4. The method according to claim 1, characterized in that, The step of retrieving multiple response results data for the first query information based on the first demand feature in the first vector knowledge base includes: Based on the first demand feature and the response vector index, a set of candidate response vectors is determined from the first vector knowledge base; wherein, the set of candidate response vectors contains at least one candidate response vector data. Calculate the cosine similarity between each candidate response vector data and the first requirement feature, and determine multiple target response vector data based on multiple cosine similarities and the similarity threshold; Based on the response data corresponding to each target response vector data, multiple response result data are determined.

5. The method according to claim 1, characterized in that, After determining the multiple response results data for the first query information, the process also includes: Determine the weight coefficient of each response result data under the preset evaluation dimensions; the preset evaluation dimensions include at least one of the following: data recommendation dimension, usage frequency dimension, and project affiliation dimension; Based on the weighting coefficients, the scoring results for each of the response result data are determined; Based on the scoring results, a preset number of response results to be displayed are determined from the multiple response result data, and the response results to be displayed are recommended for display.

6. The method according to claim 5, characterized in that, The step of recommending and displaying the response result data to be shown includes: For a single response result data to be displayed, multiple display element results corresponding to the response result data to be displayed are determined based on preset element requirements; wherein, the preset element requirements are used to indicate the display elements that need to be displayed when the response result data to be displayed is presented; Based on the results of multiple display elements and the response results to be displayed, display data in a preset display format is generated, and the display data is displayed at a preset position on the interface.

7. A question-answering recommendation system, characterized in that, The system includes a question-answering recommendation device, a computing power service device, and a first vector knowledge base; wherein: The question-and-answer recommendation device is used to acquire first information in real time. The first information includes a first question and multiple second questions. The question time corresponding to the second question is earlier than the question time corresponding to the first question. The question-answering recommendation device is further configured to send the first information to the computing power service device, so that the computing power service device performs the following operations: performing semantic analysis on the first information to determine a first demand feature; and retrieving multiple response result data for the first query information based on the first demand feature in a first vector knowledge base; wherein, the first demand feature is used to characterize the first information from at least one of the following characterization dimensions: a first characterization dimension, a second characterization dimension, and a third characterization dimension; the first characterization dimension is used to indicate the query demand content corresponding to the first information, the second characterization dimension is used to indicate the dialogue scenario to which the first information belongs, and the third characterization dimension is used to indicate the entity objects involved in the first information; the similarity between the response result data and the first demand feature is greater than a similarity threshold, and the first vector knowledge base is composed of response vector data corresponding to response data of multiple different modalities; The question-and-answer recommendation device is also used to receive multiple response result data sent by the computing power service device.

8. A question-and-answer recommendation device, characterized in that, The question-and-answer recommendation device includes: One or more processors; Storage device 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 implement the question-answering recommendation method as described in any one of claims 1-6.

9. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the question-and-answer recommendation method as described in any one of 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the question-and-answer recommendation method as described in any one of claims 1-6.

Images