B2b system intelligent customer service management and control method and system based on natural language processing

Abstract

The application discloses a B2B system intelligent customer service management and control method and system based on natural language processing, and the method comprises the following steps: acquiring natural language query information fed back by a user through multiple channels and based on a B2B platform; performing a preprocessing operation on the natural language query information to obtain a preprocessing result for the natural language query information; the preprocessing is used for converting the natural language query information into text data capable of being recognized; performing a target processing operation on the preprocessing result to obtain a target processing result for the preprocessing result; the target processing operation comprises at least one operation in a first information analysis operation based on a context understanding module, a second information analysis operation based on a semantic analysis model and a third information analysis operation based on a knowledge base; and the target processing result is used for indicating a response matter for responding to the natural language query information. It can be seen that the application can improve the intelligent degree, accuracy and humanization degree of the query information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent customer service, and in particular to a B2B system intelligent customer service management and control method and system based on natural language processing. BACKGROUND

[0002] In the existing field of B2B system intelligent customer service, with the increasing frequency of business exchanges between enterprises, users are increasingly dependent on intelligent customer service. However, the current intelligent customer service system faces many difficulties when processing user queries.

[0003] On the one hand, there are many user feedback channels, such as email, online customer service window, telephone message, etc., but the information between each channel is independent of each other, lacking an effective integration mechanism. This makes customer service personnel need to spend a lot of time and effort to collect and organize when facing a large amount of scattered information, not only low efficiency, but also easy to miss information, unable to fully understand user problems, and thus affecting service quality.

[0004] On the other hand, natural language query information has high complexity and diversity. Users have different expression habits and language styles, and the query content may involve professional knowledge and complex business logic. The existing intelligent customer service system often lacks effective preprocessing means when processing this information, making it difficult to accurately convert natural language into a data format that computers can understand and process. This leads to deviations in subsequent analysis and response links, and is unable to accurately capture user intent.

[0005] In addition, when responding to user queries, conventional intelligent customer service cannot give accurate and comprehensive responses when facing some complex or ambiguous problems, and cannot meet the diverse needs of users, reducing user satisfaction and trust in intelligent customer service. SUMMARY

[0006] The present application provides a B2B system intelligent customer service management and control method and system based on natural language processing, which can realize accurate collection of multi-channel query information, improve the intelligent degree, accuracy and humanization of query information response.

[0007] To solve the above technical problems, the first aspect of the present application discloses a B2B system intelligent customer service management and control method based on natural language processing, the method comprising:

[0008] Obtaining natural language query information fed back by users through multiple channels, the natural language query information including query information triggered by the users through a B2B platform;

[0009] performing a preprocessing operation on the natural language query information to obtain a preprocessing result corresponding to the natural language query information; the preprocessing operation is used to convert the natural language query information into text data capable of being recognized;

[0010] performing a target processing operation on the preprocessing result to obtain a target processing result corresponding to the preprocessing result; the target processing operation comprises at least one of a first information analysis operation based on a context understanding module, a second information analysis operation based on a semantic analysis model, and a third information analysis operation based on a knowledge base; and the target processing result is used to indicate a response matter responding to the natural language query information.

[0011] As an optional implementation, in the first aspect of the present application, the performing of the target processing operation on the preprocessing result to obtain the target processing result corresponding to the preprocessing result comprises:

[0012] performing a first information analysis operation on the preprocessing result according to the context understanding module to obtain a first information analysis result corresponding to the preprocessing result; the first information analysis operation comprises at least one of a historical record query, a session state analysis, and a user preference analysis;

[0013] performing a second information analysis operation on the first information analysis result according to the semantic analysis model to obtain a second information analysis result corresponding to the first information analysis result; the second information analysis operation comprises at least one of a user query intention recognition, an entity information analysis, and a user sentiment tendency analysis;

[0014] performing a third information analysis operation on the second information analysis result according to the knowledge base to obtain a third information analysis result corresponding to the second information analysis result as the target processing result; the third information analysis operation is used to match a response content responding to the second information analysis result.

[0015] As an optional implementation, in the first aspect of the present application, the performing of the first information analysis operation on the preprocessing result according to the context understanding module to obtain the first information analysis result corresponding to the preprocessing result comprises:

[0016] monitoring a session state in real time after the user triggers the natural language query information according to a preset session state model, the session state comprising any one of a session start, a session in progress, a session pause, and a session end;

[0017] When the session state is determined to be the session start or the in-process, it is detected whether there is a historical dialogue record corresponding to the preprocessing result, when it is detected that there is the historical dialogue record, information extraction is performed on the historical dialogue record according to a context understanding module, a record abstract corresponding to the historical dialogue record is obtained, and the record abstract includes at least one of historical dialogue content, question type, and solution with a high correlation degree higher than a preset correlation threshold to the natural language query information;

[0018] According to the context understanding module, user preference analysis is performed on the historical dialogue record, and query preference items, potential demand items, and interest point items of the user are obtained as user preference information;

[0019] The record abstract and the user preference information are associated and fused with the preprocessing result to update the preprocessing result, and a first information analysis result is obtained.

[0020] As an optional implementation, in the first aspect of the present application, the second information analysis operation is performed on the first information analysis result according to a semantic analysis model, and a second information analysis result corresponding to the first information analysis result is obtained, including:

[0021] According to the semantic analysis model, user query intention recognition is performed on the first information analysis result, and an intention recognition result of the user is obtained; the intention recognition result includes an intention category and a confidence score;

[0022] According to the semantic analysis model, entity information analysis is performed on the first information analysis result, and target entity information for the first information analysis result is obtained; the entity information analysis is used to extract keywords corresponding to information query in the preprocessing result; the target entity information includes one or more entity keywords;

[0023] According to the semantic analysis model, sentiment tendency analysis is performed on the first information analysis result, and target sentiment of the user triggering the natural language query information and a corresponding sentiment intensity thereof are obtained; the target sentiment includes positive sentiment or negative sentiment or neutral sentiment;

[0024] The intention recognition result, the target entity information, and the target sentiment and the corresponding sentiment intensity thereof are determined as the second information analysis result corresponding to the first information analysis result.

[0025] As an optional implementation, in the first aspect of the present application, the third information analysis operation is performed on the second information analysis result according to a knowledge base, and a third information analysis result corresponding to the second information analysis result is obtained, including:

[0026] performing a first-level matching operation on the second information analysis result according to the knowledge base, to obtain a first-level matching result corresponding to the second information analysis result; the first-level matching operation comprises knowledge item retrieval; and the first-level matching result comprises at least one knowledge item to be screened;

[0027] performing a second-level matching operation on the first-level matching result according to the knowledge base, to obtain a second-level matching result corresponding to the first-level matching result; the second-level matching operation comprises deep analysis, matching degree calculation, applicability scoring and sorting based on the first information analysis result and the second information analysis result; and the second-level matching result comprises a target knowledge item meeting a screening requirement;

[0028] performing information filling and sentiment adjustment on the second-level matching result according to the second information analysis result, to obtain a target output result as a third information analysis result.

[0029] As an optional implementation form, in the first aspect of the present application, the second-level matching operation performed on the first-level matching result according to the knowledge base to obtain a second-level matching result corresponding to the first-level matching result comprises:

[0030] for each of the knowledge items to be screened, performing a text similarity calculation operation on the second information analysis result and the knowledge item to be screened according to the knowledge base, to obtain a similarity calculation result between the second information analysis result and the knowledge item to be screened; the text similarity calculation operation comprises text preprocessing, feature extraction and similarity calculation;

[0031] performing a semantic annotation operation on the second information analysis result and the knowledge item to be screened according to the knowledge base, to obtain a semantic annotation result corresponding to the second information analysis result and the knowledge item to be screened; the semantic annotation operation comprises syntax structure analysis, semantic role labeling and semantic role comparison;

[0032] calculating a comprehensive similarity score between the second information analysis result and the knowledge item to be screened, and calculating an applicability score of the knowledge item to be screened corresponding to the query requirement and / or the conversation scenario of the user, according to the similarity calculation result and the semantic annotation result corresponding to the knowledge item to be screened;

[0033] performing scoring and sorting on all the knowledge items to be screened, taking the comprehensive similarity score and the applicability score corresponding to each of the knowledge items to be screened as a reference, and selecting any of the knowledge items to be screened with a ranking within a preset ranking range from the corresponding scoring and sorting result as a target knowledge item, as a second-level matching result.

[0034] As an optional implementation, in the first aspect of the present application, the text preprocessing operation includes word segmentation, stop word removal, stem extraction, and phrase reorganization; and the text preprocessing is used to process the second information analysis result into a standardized text representation.

[0035] The semantic role labeling operation is used to label all semantic roles corresponding to the second information analysis result, and all the semantic roles include agent roles, patient roles, tools, times, and locations.

[0036] The second aspect of the present application discloses a B2B system intelligent customer service management and control system based on natural language processing, which comprises:

[0037] An acquisition module is configured to acquire natural language query information fed back by a user through multiple channels, wherein the natural language query information includes query information triggered by the user through a B2B platform.

[0038] A preprocessing module is configured to perform a preprocessing operation on the natural language query information to obtain a preprocessing result for the natural language query information; and the preprocessing is used to convert the natural language query information into text data that can be recognized.

[0039] A target processing module is configured to perform a target processing operation on the preprocessing result to obtain a target processing result for the preprocessing result; the target processing operation includes at least one of a first information analysis operation based on a context understanding module, a second information analysis operation based on a semantic analysis model, and a third information analysis operation based on a knowledge base; and the target processing result is used to indicate a response matter for responding to the natural language query information.

[0040] As an optional implementation, in the second aspect of the present application, the target processing module comprises:

[0041] A first information analysis submodule is configured to perform a first information analysis operation on the preprocessing result according to a context understanding module to obtain a first information analysis result corresponding to the preprocessing result; and the first information analysis operation includes at least one of historical record query, session state analysis, and user preference analysis.

[0042] A second information analysis submodule is configured to perform a second information analysis operation on the first information analysis result according to a semantic analysis model to obtain a second information analysis result corresponding to the first information analysis result; and the second information analysis operation includes at least one of user query intent recognition, entity information analysis, and user sentiment tendency analysis.

[0043] The third information analysis submodule is configured to perform a third information analysis operation on the second information analysis result according to the knowledge base, to obtain a third information analysis result corresponding to the second information analysis result as a target processing result; and the third information analysis operation is configured to match a response content corresponding to the second information analysis result.

[0044] As an optional implementation, in the second aspect, the first information analysis submodule is configured to perform a first information analysis operation on the preprocessing result according to the context understanding module, to obtain a first information analysis result corresponding to the preprocessing result, and the manner specifically includes:

[0045] The conversation state after the user triggers the natural language query information is monitored in real time according to a preset conversation state model, and the conversation state includes any one of a conversation start, a conversation in progress, a conversation pause, and a conversation end;

[0046] When the conversation state is determined to be the conversation start or the conversation in progress, it is detected whether there is a historical dialogue record corresponding to the preprocessing result, when it is detected that there is the historical dialogue record, information extraction is performed on the historical dialogue record according to the context understanding module, to obtain a record summary corresponding to the historical dialogue record, and the record summary includes at least one of historical dialogue content, a question type, and a solution that has a high correlation with the natural language query information and is higher than a preset correlation threshold;

[0047] User preference analysis is performed on the historical dialogue record according to the context understanding module, to obtain a query preference item, a potential demand item, and an interest point item of the user as user preference information;

[0048] The record summary and the user preference information are associated and fused with the preprocessing result, to update the preprocessing result, to obtain a first information analysis result.

[0049] As an optional implementation, in the second aspect, the second information analysis submodule is configured to perform a second information analysis operation on the first information analysis result according to a semantic analysis model, to obtain a second information analysis result corresponding to the first information analysis result, and the manner specifically includes:

[0050] User query intent recognition is performed on the first information analysis result according to a semantic analysis model, to obtain an intent recognition result of the user; and the intent recognition result includes an intent category and a confidence score;

[0051] According to the semantic analysis model, entity information analysis is performed on the first information analysis result to obtain target entity information corresponding to the first information analysis result; the entity information analysis is used to extract keywords corresponding to the information query in the preprocessing result; and the target entity information includes one or more entity keywords.

[0052] According to the semantic analysis model, sentiment tendency analysis is performed on the first information analysis result to obtain a target sentiment of the user triggering the natural language query information and a corresponding sentiment intensity; and the target sentiment includes positive sentiment, negative sentiment or neutral sentiment.

[0053] The intent recognition result, the target entity information, and the target sentiment and the corresponding sentiment intensity are determined as second information analysis results corresponding to the first information analysis result.

[0054] As an optional implementation, in the second aspect of the present application, the third information analysis sub-module performs third information analysis operation on the second information analysis result according to the knowledge base to obtain a third information analysis result corresponding to the second information analysis result, and the manner specifically includes:

[0055] According to the knowledge base, first-level matching operation is performed on the second information analysis result to obtain a first-level matching result corresponding to the second information analysis result; the first-level matching operation includes knowledge item retrieval; and the first-level matching result includes at least one knowledge item to be screened.

[0056] According to the knowledge base, second-level matching operation is performed on the first-level matching result to obtain a second-level matching result corresponding to the first-level matching result; the second-level matching operation includes deep analysis, matching degree calculation, applicability scoring and sorting based on the first information analysis result and the second information analysis result; and the second-level matching result includes a target knowledge item meeting the screening requirement.

[0057] According to the second information analysis result, information filling and sentiment adjustment are performed on the second-level matching result to obtain a target output result as a third information analysis result.

[0058] As an optional implementation, in the second aspect of the present application, the third information analysis sub-module performs second-level matching operation on the first-level matching result according to the knowledge base to obtain a second-level matching result corresponding to the first-level matching result, and the manner specifically includes:

[0059] For each of the to-be-screened knowledge items, a text similarity calculation operation is performed on the second information analysis result and the to-be-screened knowledge item according to the knowledge base, to obtain a similarity calculation result between the second information analysis result and the to-be-screened knowledge item; the text similarity calculation operation includes a text preprocessing operation, a feature extraction operation, and a similarity calculation operation.

[0060] A semantic labeling operation is performed on the second information analysis result and the to-be-screened knowledge item according to the knowledge base, to obtain a semantic labeling result corresponding to the second information analysis result and the to-be-screened knowledge item; the semantic labeling operation includes a syntax structure parsing operation, a semantic role labeling operation, and a semantic role comparison operation.

[0061] According to the similarity calculation result and the semantic labeling result corresponding to the to-be-screened knowledge item, a comprehensive similarity score between the second information analysis result and the to-be-screened knowledge item is calculated, and an applicability score of the to-be-screened knowledge item corresponding to the query requirement and / or the conversation scenario of the user is calculated.

[0062] Based on the comprehensive similarity score and the applicability score corresponding to each of the to-be-screened knowledge items, a score ranking is performed on all the to-be-screened knowledge items, and any to-be-screened knowledge item ranked within a preset ranking range is selected as a target knowledge item from the corresponding score ranking result, and is taken as a secondary matching result.

[0063] As an optional implementation, in the second aspect of the application, the text preprocessing operation includes word segmentation, stop word removal, stem extraction, and word group reorganization; the text preprocessing is used to process the second information analysis result into a standardized text representation.

[0064] The semantic role labeling operation is used to label all semantic roles corresponding to the second information analysis result, and all the semantic roles include an agent role, a patient role, a tool, a time, and a place.

[0065] The third aspect of the application discloses a B2B system intelligent customer service management and control device based on natural language processing, which comprises:

[0066] A memory storing executable program codes;

[0067] A processor coupled with the memory;

[0068] The processor invokes the executable program codes stored in the memory to execute the B2B system intelligent customer service management and control method based on natural language processing disclosed in the first aspect of the application.

[0069] The fourth aspect of the present application discloses a computer storage medium, the computer storage medium stores computer instructions, when the computer instructions are called, the computer instructions are used to execute the intelligent customer service management and control method of the B2B system based on natural language processing disclosed in the first aspect of the present application.

[0070] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0071] In the embodiments of the present application, a B2B system intelligent customer service management and control method based on natural language processing is provided, which comprises: obtaining natural language query information fed back by a user through multiple channels, the natural language query information comprising query information triggered by the user through a B2B platform; performing a preprocessing operation on the natural language query information to obtain a preprocessing result for the natural language query information; the preprocessing is used to convert the natural language query information into text data that can be recognized; performing a target processing operation on the preprocessing result to obtain a target processing result for the preprocessing result; the target processing operation comprises at least one of a first information analysis operation based on a context understanding module, a second information analysis operation based on a semantic analysis model, and a third information analysis operation based on a knowledge base; the target processing result is used to indicate a response matter for responding to the natural language query information. As can be seen, by performing a preprocessing operation on the natural language query information to convert it into standardized and structured text data, the present application can effectively remove redundant data and improve the processing efficiency and accuracy of the natural language query information; and by combining the context understanding module, the semantic analysis model, and the knowledge base, the preprocessing result is analyzed in multiple dimensions and at a deep level, so that the user's query intention can be accurately captured and the user's potential demand can be mined, which is beneficial to improve the response accuracy and reliability of the subsequent natural language query information triggered by the user and improve the user's query experience. BRIEF DESCRIPTION OF DRAWINGS

[0072] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0073] Figure 1 is a flowchart of a B2B system intelligent customer service management and control method based on natural language processing disclosed by the embodiments of the present application;

[0074] Figure 2 is a flowchart of another B2B system intelligent customer service management and control method based on natural language processing disclosed by the embodiments of the present application;

[0075] Figure 3is a structural schematic view of a B2B system intelligent customer service management and control system based on natural language processing disclosed by an embodiment of the application.

[0076] Figure 4 is a structural schematic view of another B2B system intelligent customer service management and control system based on natural language processing disclosed by an embodiment of the application.

[0077] Figure 5 is a structural schematic view of a B2B system intelligent customer service management and control device based on natural language processing disclosed by an embodiment of the application. DETAILED DESCRIPTION

[0078] In order to enable persons skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by persons skilled in the art without creative labor fall within the scope of protection of the present application.

[0079] The terms "first", "second", and the like in the specification of the present application and the above-mentioned drawings are used to distinguish different objects, and are not used to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product, or end including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product, or end.

[0080] In this document, the term "embodiment" means that the specific features, structures, or characteristics described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor is it independent or alternative to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0081] The application discloses a B2B system intelligent customer service management and control method and system based on natural language processing.

[0082] Embodiment one

[0083] Please refer to Figure 1 , Figure 1 is a flowchart of a B2B system intelligent customer service management and control method based on natural language processing disclosed by the embodiment of the application. Wherein, Figure 1 The B2B system intelligent customer service management and control method based on natural language processing described above can be applied to a B2B system, and the B2B system can be a B2B system intelligent customer service management and control system based on natural language processing. The embodiment of the application is not limited. As shown in Figure 1 The B2B system intelligent customer service management and control method based on natural language processing can include the following operations:

[0084] 101, obtaining natural language query information fed back by a user through multiple channels, wherein the natural language query information includes query information triggered by the user through a B2B platform.

[0085] In the embodiment of the application, the multiple channels can include online channels and offline channels. When it is an online channel, the triggering / submitting of the natural language query information can be performed in the form of an online customer service window, a message board function and a submission form on the B2B platform. The triggering / submitting of the natural language query information can also be performed through an official social media account, an industry forum and community or an email. When it is an offline channel, the triggering / submitting of the natural language query information can be performed in the form of a customer service telephone, a questionnaire filled in at an offline industry exhibition or a seminar, and the like. The inquiry record of the customer service telephone on the B2B platform is collected, and the uploaded questionnaire data is read to obtain the natural language query information.

[0086] In the embodiment of the application, the natural language query information refers to non-machine language triggered by a user. For example, "I want to inquire about the ab function on your platform", or "What is the cd function of your platform" and the like.

[0087] In the embodiment of the present application, the natural language query information triggered by the user is acquired through multiple channels, so that various query information triggered by the user through the B2B platform can be covered, and the acquisition integrity and timeliness of the natural language query information are improved.

[0088] 102. performing a preprocessing operation on the natural language query information to obtain a preprocessing result for the natural language query information; the preprocessing is used to convert the natural language query information into text data that can be recognized.

[0089] In the embodiment of the present application, the preprocessing operation includes text vectorization, and at least one of text cleaning, part-of-speech processing, part-of-speech tagging, and spelling correction.

[0090] In the embodiment of the present application, the text vectorization is used to convert the natural language query information into vector information that can be processed / recognized by a computer; and the text vectorization can be performed by using a conventional Bag of Words model or a TF-IDF (Term Frequency-Inverse Document Frequency) model.

[0091] In the embodiment of the present application, the text cleaning is used to filter out noise in the natural language query information, such as HTML tags (if the information comes from a web form), special characters (such as emoticons, garbled characters), redundant spaces and line breaks, etc. The text cleaning is also used for text format unification, such as unification of full-width characters and half-width characters, unification of uppercase and lowercase English, etc.

[0092] In the embodiment of the present application, the word segmentation processing includes dictionary-based word segmentation and / or word segmentation combined with a statistical method; the dictionary-based word segmentation divides the text according to certain rules by using a pre-constructed word segmentation dictionary. For example, for the Chinese text “I want to query order logistics information”, the word segmentation dictionary contains the words “I”, “want”, “query”, “order”, “logistics”, and “information”, and the text can be divided into “I / want / query / order / logistics / information” through dictionary matching. The word segmentation combined with the statistical method can be statistical word segmentation based on an n-gram model, specifically, whether word segmentation should be performed is determined by analyzing the appearance frequency of adjacent words or characters in the text. For example, for “Nanjing Yangtze River Bridge”, it is found through statistics that the appearance frequency of “Nanjing City” and “Yangtze River Bridge” is high, while the appearance frequency of combinations such as “Nanjing City” and “Yangtze River Bridge” is low, so the text can be correctly divided into “Nanjing City / Yangtze River Bridge”.

[0093] In the embodiments of the present invention, the part-of-speech tagging can, on the basis of the above-mentioned part-of-speech processing, further tag the results of the part-of-speech processing, including tagging nouns, verbs, and adjectives. For example, for the segmented text "我 / 想 / 查询 / 订单 / 物流 / 信息", "我" is tagged as a personal pronoun, "想" is tagged as a verb, "查询" is tagged as a verb, "订单" is tagged as a noun, "物流" is tagged as a noun, and "信息" is tagged as a noun.

[0094] In the embodiments of the present invention, the spelling correction can use an algorithm based on the edit distance to calculate the edit distance between the word input by the user and each word in the dictionary, and select the word with the smallest edit distance as the corrected result. For example, when the user inputs "订丹" (which may be a spelling mistake of "订单"), by comparing it with the words in the dictionary, it is found that the edit distance between "订单" and "订丹" is the smallest, so "订丹" is corrected to "订单".

[0095] 103. Perform a target processing operation on the preprocessing result to obtain a target processing result for the preprocessing result; the target processing operation includes at least one of the first information analysis operation based on the context understanding module, the second information analysis operation based on the semantic analysis model, and the third information analysis operation based on the knowledge base; the target processing result is used to indicate the response item for the natural language query information.

[0096] It can be seen that implementing Figure 1 the described intelligent customer service control method for the B2B system based on natural language processing, by performing preprocessing operations on the natural language query information and converting it into standardized and structured text data, can effectively remove redundant data and improve the processing efficiency and accuracy for the natural language query information; and, by combining the context understanding module, the semantic analysis model, and the knowledge base, performing multi-dimensional and in-depth information analysis on the preprocessing result, it is possible to accurately capture the user's query intention,挖掘用户的潜在需求,有利于提高后续针对用户触发的自然语言查询信息的响应准确性与可靠性,提高用户的查询体验.

[0097] Embodiment 2

[0098] Please refer to Figure 2 , Figure 2 which is a schematic flowchart of another intelligent customer service control method for the B2B system based on natural language processing disclosed in the embodiments of the present invention. Among them, Figure 2 the described intelligent customer service control method for the B2B system based on natural language processing can be applied to the B2B system, and the B2B system can further be an intelligent customer service control system for the B2B system based on natural language processing, which is not limited in the embodiments of the present invention. As Figure 2As shown, the B2B system intelligent customer service management and control method based on natural language processing can include the following operations:

[0099] 201, acquire natural language query information fed back by a user through multiple channels, the natural language query information including query information triggered by the user through a B2B platform.

[0100] 202, perform a preprocessing operation on the natural language query information to obtain a preprocessing result for the natural language query information; the preprocessing is used to convert the natural language query information into text data that can be recognized.

[0101] In the embodiment of the application, for other descriptions of steps 201-202, please refer to other specific descriptions of steps 101-102 in Embodiment One, and the embodiment of the application will not be repeated here.

[0102] 203, according to the context understanding module, perform a first information analysis operation on the preprocessing result to obtain a first information analysis result corresponding to the preprocessing result; the first information analysis operation includes at least one operation of historical record query, session state analysis and user preference analysis.

[0103] 204, according to a semantic analysis model, perform a second information analysis operation on the first information analysis result to obtain a second information analysis result corresponding to the first information analysis result; the second information analysis operation includes at least one operation of user query intent recognition, entity information analysis and user sentiment tendency analysis.

[0104] 205, according to a knowledge base, perform a third information analysis operation on the second information analysis result to obtain a third information analysis result corresponding to the second information analysis result as a target processing result; the third information analysis operation is used to match a response content responding to the second information analysis result.

[0105] In the embodiment of the application, the response content can be based on a preset response template, and the third information analysis result is filled in the response template and displayed.

[0106] It can be seen that the implementation Figure 2The described natural language processing-based B2B system intelligent customer service management and control method realizes efficient and accurate response of the B2B system to user queries through a multi-level information analysis and processing flow. Specifically, the context understanding module can be used to combine historical record queries, session state analysis and user preference analysis to comprehensively capture the context information of user queries, which is conducive to improving the association degree of information analysis and context and then improving the accuracy of information analysis. With the help of a semantic analysis model, the user query intent, entity information and sentiment tendency can be deeply identified and analyzed, which is conducive to improving the analysis accuracy of user real query demand. The analysis result can be intelligently matched based on a knowledge base to quickly locate and generate response content highly consistent with user demand, thereby realizing efficient use of knowledge resources, improving the intelligent degree of the response process corresponding to user query demand, and improving service efficiency and quality.

[0107] In an optional embodiment, the manner of performing, by the context understanding module, a first information analysis operation on the preprocessing result to obtain a first information analysis result corresponding to the preprocessing result specifically includes:

[0108] According to a preset session state model, the session state after the user triggers the natural language query information is monitored in real time, and the session state includes any one of a session start, a session in progress, a session pause, and a session end;

[0109] When it is determined that the session state is a session start or a session in progress, it is detected whether there is a historical dialogue record corresponding to the preprocessing result. When it is detected that there is a historical dialogue record, information extraction is performed on the historical dialogue record according to the context understanding module to obtain a record abstract corresponding to the historical dialogue record. The record abstract includes at least one of historical dialogue content, a question type, and a solution that has a higher correlation with the natural language query information than a preset correlation threshold.

[0110] According to the context understanding module, user preference analysis is performed on the historical dialogue record to obtain a query preference item, a potential demand item, and an interest point item of the user as user preference information.

[0111] The record abstract and the user preference information are associated and fused with the preprocessing result to update the preprocessing result to obtain the first information analysis result.

[0112] It can be seen that in the optional embodiment, based on the preset conversation state model, the real-time monitoring of the whole cycle of the user conversation can be realized, the key nodes such as the start, the process, the pause and the end of the conversation can be accurately identified, the dynamic context benchmark for subsequent information analysis is provided, and the analysis process is highly synchronized with the conversation process; and in the active stage (start / process) of the conversation, the historical dialogue records can be automatically retrieved and associated, the high-correlation degree abstract information (including historical dialogue content, question type, solution, etc.) can be extracted through the context understanding module, so that the historical interaction experience is effectively reused, and the efficiency of the current query is improved; and multi-dimensional preference analysis can also be performed based on the historical dialogue records, the user query preference, the potential demand and the interest point portrait are accurately constructed, the depth of the personalized service capability is strengthened, and the matching degree and the accuracy of the response content and the real demand of the user are further improved.

[0113] In another optional embodiment, the manner of performing the second information analysis operation on the first information analysis result according to the semantic analysis model to obtain the second information analysis result corresponding to the first information analysis result specifically includes:

[0114] According to the semantic analysis model, the user query intention recognition is performed on the first information analysis result to obtain an intention recognition result of the user; the intention recognition result includes an intention category and a confidence score;

[0115] According to the semantic analysis model, the entity information analysis is performed on the first information analysis result to obtain target entity information corresponding to the first information analysis result; the entity information analysis is used to extract keywords corresponding to the information query in the preprocessing result; the target entity information includes one or more entity keywords;

[0116] According to the semantic analysis model, the sentiment tendency analysis is performed on the first information analysis result to obtain a target sentiment of the user triggering the natural language query information and a corresponding sentiment intensity thereof; the target sentiment includes a positive sentiment or a negative sentiment or a neutral sentiment;

[0117] The intention recognition result, the target entity information and the target sentiment and the corresponding sentiment intensity thereof are determined as the second information analysis result corresponding to the first information analysis result.

[0118] In the optional embodiment, the entity information analysis is also used to extract other entity information corresponding to the user, the other entity information includes a user id, a product name corresponding to a product currently queried by the user, a manufacturer name corresponding to the product name, geographical information, time information and the like, wherein the geographical information can be information corresponding to a product delivery place or a production place; the time information can be product production time or supply / delivery time and the like.

[0119] It can be seen that in the optional embodiment, the semantic analysis model is used to perform multi-dimensional intent recognition on the first information analysis result, and output a structured result containing an intent category and a confidence score. The mechanism can improve the positioning accuracy of the core demand of the user query. The semantic analysis model can be used to perform entity information extraction on the preprocessing result, automatically identify and output target entity keywords highly related to the query, realize efficient conversion from unstructured text to structured data, and improve the efficiency and accuracy of subsequent knowledge base retrieval. In addition, the analysis module of the sentiment tendency can be introduced to accurately identify the sentiment tendency implied in the current query of the user, and perform quantitative analysis of the corresponding sentiment intensity value, realize accurate perception of the emotional state of the user, and provide data support for subsequent personalized and humanized response output.

[0120] In the optional embodiment, further, the manner of performing entity information analysis on the first information analysis result according to the semantic analysis model includes:

[0121] performing first entity analysis on the first information analysis result according to the semantic analysis model to obtain a first entity analysis result corresponding to the first information analysis result; the first entity analysis is used to perform preliminary entity extraction on the first information analysis result;

[0122] determining whether the first entity analysis result meets a disambiguation fusion condition, and if so, performing a disambiguation fusion processing operation on the first entity analysis result to obtain a disambiguation fusion processing result corresponding to the first entity analysis result; the disambiguation fusion processing operation is used to disambiguate entities with ambiguity and / or to fuse different entities with the same meaning; for example, for the entity "apple", it is determined according to the context whether it refers to a fruit or Apple Inc.; for another example, "Apple Inc." and "iPhone" are identified as entities related to the same theme and are associated;

[0123] performing second entity analysis on the disambiguation fusion processing result according to the semantic analysis model to obtain a second entity analysis result corresponding to the disambiguation fusion processing result as the target entity information for the first information analysis result; the second entity analysis includes key entity screening and entity information sorting.

[0124] For example, if the user query is product information, the key entity screening is used to adjust the focus on product names, brand names, model numbers, etc.

[0125] Optionally, after obtaining the second entity analysis result corresponding to the disambiguation fusion processing result, sample verification is performed on the second entity analysis result according to the semantic analysis model to obtain a sample verification result corresponding to the second entity analysis result, and the sample verification result includes entity recognition accuracy and completeness.

[0126] When the sampling verification result indicates that the entity recognition accuracy is lower than the preset accuracy threshold and / or the completeness is lower than the preset completeness threshold, model optimization is performed on the semantic analysis model according to the sampling verification result.

[0127] It can be seen that in the optional embodiment, by performing a multi-level resolution operation on the first information analysis result, including preliminary entity extraction, disambiguation fusion, deep key entity extraction and sorting, the fineness and accuracy of the entity information resolution are improved.

[0128] In yet another optional embodiment, the above-mentioned manner of performing a third information analysis operation on the second information analysis result according to the knowledge base to obtain a third information analysis result corresponding to the second information analysis result specifically comprises:

[0129] performing a first-level matching operation on the second information analysis result according to the knowledge base to obtain a first-level matching result corresponding to the second information analysis result; the first-level matching operation comprises knowledge item retrieval; the first-level matching result comprises at least one knowledge item to be screened;

[0130] performing a second-level matching operation on the first-level matching result according to the knowledge base to obtain a second-level matching result corresponding to the first-level matching result; the second-level matching operation comprises deep analysis, matching degree calculation, applicability scoring and sorting based on the first information analysis result and the second information analysis result; the second-level matching result comprises a target knowledge item meeting the screening requirements;

[0131] performing information filling and sentiment adjustment on the second-level matching result according to the second information analysis result to obtain a target output result as the third information analysis result.

[0132] In the optional embodiment, the first-level matching operation can comprise knowledge item retrieval and / or response template retrieval; correspondingly, the matching result comprises a knowledge item to be screened and / or a response template to be screened.

[0133] It can be seen that in the optional embodiment, a two-stage matching process is designed, which integrates the first information analysis result (context feature) and the second information analysis result (semantic feature), and combines multi-dimensional analysis (including matching degree calculation, applicability score and dynamic sorting) to realize step-by-step and accurate screening of knowledge items. The problem of response deviation caused by single matching dimension is solved, which is conducive to improving the accuracy and applicability of the target knowledge item obtained by screening; and for the two-stage matching result, a corresponding information filling and emotion adjustment mechanism is further set to realize intelligent adjustment between standardized knowledge items and personalized response content, solve the problems of low matching efficiency and mechanical response in traditional knowledge base application, improve the humanization degree of the output third information analysis result / overall intelligent customer system, thereby improving the user's interactive experience.

[0134] In another optional embodiment, the above-mentioned manner of performing a two-stage matching operation on the first matching result according to the knowledge base to obtain a two-stage matching result corresponding to the first matching result specifically comprises:

[0135] For each knowledge item to be screened, a text similarity calculation operation is performed on the second information analysis result and the knowledge item to be screened according to the knowledge base to obtain a similarity calculation result between the second information analysis result and the knowledge item to be screened; the text similarity calculation operation includes a text preprocessing operation, a feature extraction operation and a similarity calculation operation;

[0136] A semantic annotation operation is performed on the second information analysis result and the knowledge item to be screened according to the knowledge base to obtain a semantic annotation result corresponding to the second information analysis result and the knowledge item to be screened; the semantic annotation operation includes a syntactic structure analysis operation, a semantic role labeling operation and a semantic role comparison operation;

[0137] According to the similarity calculation result and the semantic annotation result corresponding to the knowledge item to be screened, a comprehensive similarity score between the second information analysis result and the knowledge item to be screened is calculated, and an applicability score of the knowledge item to be screened and the user's query demand and / or the conversation scene is calculated;

[0138] Taking the comprehensive similarity score and the applicability score corresponding to each knowledge item to be screened as a reference, a scoring and sorting operation is performed on all knowledge items to be screened, and any knowledge item to be screened within a predetermined ranking range is selected from the corresponding scoring and sorting result as a target knowledge item, and as a two-stage matching result.

[0139] In the optional embodiment, the text preprocessing operation includes word segmentation, stop word removal, stem extraction and word group reorganization; the text preprocessing is used to process the second information analysis result into a standardized text representation;

[0140] The semantic role labeling operation is used for labeling all semantic roles corresponding to the second information analysis result, all semantic roles including an agent role, a patient role, a tool, time and place.

[0141] In the optional embodiment, the agent role is the execution subject of the matter, for example, a purchaser; and the patient role is the executed subject of the matter, for example, a supplier winning the bet with the purchaser.

[0142] In the optional embodiment, the text similarity calculation operation can be implemented based on a pre-constructed deep learning model, and further, the deep learning model is specialized for multi-dimensional text similarity calculation and evaluation; the deep learning model includes a standardization preprocessing process based on word segmentation, stop word removal, stem extraction and word group reorganization, and also involves conventional text feature extraction and similarity calculation. Through the deep learning model, the expression differences of natural language can be effectively eliminated, and the semantic closeness and similarity between the second information analysis result and the knowledge item can be accurately quantified, thereby providing data basis for subsequent knowledge item screening.

[0143] It can be seen that, in the optional embodiment, by constructing a three-level screening mechanism corresponding to "text similarity-semantic role-bi-dimensional score", the optimal response content is quickly located from a large amount of knowledge items, and the problems of semantic gap and low scene adaptation in traditional knowledge matching are effectively solved, the service accuracy and adaptability of the B2B intelligent customer service system are improved, and the interactive experience and business processing efficiency of users using the B2B intelligent customer service system are improved.

[0144] Embodiment Three

[0145] Please refer to Figure 3 , Figure 3 is a structural schematic diagram of a B2B system intelligent customer service management and control system based on natural language processing disclosed by the embodiment of the application. As shown in Figure 3 , the B2B system intelligent customer service management and control system based on natural language processing can include an acquisition module 301, a preprocessing module 302 and a target processing module 303, wherein:

[0146] The acquisition module 301 is configured to acquire natural language query information fed back by a user through multiple channels, wherein the natural language query information includes query information triggered by the user through a B2B platform.

[0147] The preprocessing module 302 is configured to perform a preprocessing operation on the natural language query information to obtain a preprocessing result for the natural language query information; and the preprocessing is used to convert the natural language query information into text data that can be recognized.

[0148] The target processing module 303 is configured to perform a target processing operation on the pre-processing result to obtain a target processing result corresponding to the pre-processing result, wherein the target processing operation comprises at least one of a first information analysis operation based on the context understanding module, a second information analysis operation based on the semantic analysis model, and a third information analysis operation based on the knowledge base, and the target processing result is used to indicate a response matter responding to the natural language query information.

[0149] It can be seen that the implementation of the B2B system intelligent customer service management and control system based on natural language processing described above can effectively remove redundant data and improve the processing efficiency and accuracy of the natural language query information by performing pre-processing on the natural language query information to convert it into standardized and structured text data. Figure 3 In addition, by combining the context understanding module, the semantic analysis model and the knowledge base, the pre-processing result is analyzed in multiple dimensions and at a deep level, so as to accurately capture the query intention of the user and mine the potential needs of the user, which is conducive to improving the response accuracy and reliability of the subsequent natural language query information triggered by the user and improving the query experience of the user.

[0150] In an optional embodiment, please refer to Figure 4 , Figure 4 is another structure diagram of the B2B system intelligent customer service management and control system based on natural language processing disclosed by the embodiment of the present application. As Figure 4 shown, the target processing module 303 comprises:

[0151] The first information analysis submodule 3031 is configured to perform a first information analysis operation on the pre-processing result according to the context understanding module to obtain a first information analysis result corresponding to the pre-processing result, wherein the first information analysis operation comprises at least one of a historical record query, a session state analysis and a user preference analysis;

[0152] The second information analysis submodule 3032 is configured to perform a second information analysis operation on the first information analysis result according to the semantic analysis model to obtain a second information analysis result corresponding to the first information analysis result, wherein the second information analysis operation comprises at least one of a user query intention recognition, an entity information analysis and a user sentiment analysis;

[0153] The third information analysis submodule 3033 is configured to perform a third information analysis operation on the second information analysis result according to the knowledge base to obtain a third information analysis result corresponding to the second information analysis result as the target processing result, wherein the third information analysis operation is used to match a response content responding to the second information analysis result.

[0154] It can be seen that in the optional embodiment, through the multi-level information analysis processing flow, efficient and accurate response of the B2B system to the user query is realized. Specifically, the context understanding module can be used to combine historical record query, session state analysis and user preference analysis to comprehensively capture the context information of the user query, which is conducive to improving the association degree of information analysis and context and then improving the accuracy of information analysis; the semantic analysis model can be used to deeply identify and analyze the user query intention, entity information and sentiment tendency, which is conducive to improving the analysis accuracy of the real query demand of the user; the knowledge base can be used to intelligently match the analysis result, quickly locate and generate response content highly consistent with the user demand, realize efficient use of knowledge resources, and improve the intelligent degree of the response process corresponding to the user query demand and improve the service efficiency and quality.

[0155] In another optional embodiment, the first information analysis submodule 3031 performs a first information analysis operation on the preprocessing result according to the context understanding module, and the manner of obtaining the first information analysis result corresponding to the preprocessing result specifically includes:

[0156] According to the preset session state model, the session state after the user triggers the natural language query information is monitored in real time, and the session state includes any one of a session start, a session in progress, a session pause, and a session end;

[0157] When it is determined that the session state is a session start or a session in progress, it is detected whether there is a historical dialogue record corresponding to the preprocessing result, and when it is detected that there is a historical dialogue record, information extraction is performed on the historical dialogue record according to the context understanding module to obtain a record summary corresponding to the historical dialogue record, and the record summary includes at least one of historical dialogue content, a question type, and a solution with a high correlation degree with the natural language query information higher than a preset correlation threshold;

[0158] According to the context understanding module, user preference analysis is performed on the historical dialogue record to obtain a query preference item, a potential demand item and an interest point item of the user as user preference information;

[0159] The record summary and the user preference information are associated and fused with the preprocessing result to update the preprocessing result to obtain the first information analysis result.

[0160] It can be seen that in the optional embodiment, based on the preset conversation state model, real-time monitoring of the entire cycle of the user conversation can be realized, and key nodes such as conversation start, in progress, pause, end, etc. are accurately identified, which provides a dynamic context benchmark for subsequent information analysis, and ensures that the analysis process is highly synchronized with the conversation process; and in the active stage of the conversation (start / in progress), historical dialogue records can be automatically retrieved and associated, high-correlation summary information (including historical dialogue content, question type, solution, etc.) can be extracted through the context understanding module, thereby effectively reusing historical interaction experience and improving the efficiency of the current query; and multi-dimensional preference analysis can also be performed based on the historical dialogue records to accurately construct user query preferences, potential needs and interest point portraits, realize deep reinforcement of personalized service capabilities, and further improve the matching degree and accuracy of the response content and the user's real needs.

[0161] In yet another optional embodiment, the second information analysis submodule 3032 performs a second information analysis operation on the first information analysis result according to the semantic analysis model, and the manner in which the second information analysis result corresponding to the first information analysis result is obtained specifically includes:

[0162] According to the semantic analysis model, user query intent recognition is performed on the first information analysis result to obtain an intent recognition result of the user; the intent recognition result includes an intent category and a confidence score;

[0163] According to the semantic analysis model, entity information analysis is performed on the first information analysis result to obtain target entity information corresponding to the first information analysis result; the entity information analysis is used to extract keywords corresponding to the information query in the preprocessing result; the target entity information includes one or more entity keywords;

[0164] According to the semantic analysis model, sentiment analysis is performed on the first information analysis result to obtain a target sentiment of the user triggering the natural language query information and a corresponding sentiment intensity; the target sentiment includes a positive sentiment or a negative sentiment or a neutral sentiment;

[0165] The intent recognition result, the target entity information, and the target sentiment and the corresponding sentiment intensity are determined as the second information analysis result corresponding to the first information analysis result.

[0166] It can be seen that in the optional embodiment, the semantic analysis model is used to perform multi-dimensional intent recognition on the first information analysis result, and output a structured result containing an intent category and a confidence score. The mechanism can improve the positioning accuracy of the core demand of the user query. The semantic analysis model can be used to perform entity information extraction on the preprocessed result, automatically identify and output target entity keywords highly related to the query, realize efficient conversion from unstructured text to structured data, and improve the efficiency and accuracy of subsequent knowledge base retrieval. In addition, the sentiment analysis module can be introduced to accurately identify the sentiment of the current query and quantify the corresponding sentiment intensity value, accurately perceive the user's emotional state, and provide data support for subsequent personalized and humanized response output.

[0167] In another optional embodiment, the third information analysis submodule 3033 performs a third information analysis operation on the second information analysis result according to the knowledge base to obtain a third information analysis result corresponding to the second information analysis result. The third information analysis operation includes the following steps:

[0168] According to the knowledge base, a first-level matching operation is performed on the second information analysis result to obtain a first-level matching result corresponding to the second information analysis result. The first-level matching operation includes knowledge item retrieval. The first-level matching result includes at least one knowledge item to be screened.

[0169] According to the knowledge base, a second-level matching operation is performed on the first-level matching result to obtain a second-level matching result corresponding to the first-level matching result. The second-level matching operation includes deep analysis, matching degree calculation, applicability scoring and sorting based on the first information analysis result and the second information analysis result. The second-level matching result includes a target knowledge item meeting the screening requirements.

[0170] According to the second information analysis result, information filling and sentiment adjustment are performed on the second-level matching result to obtain a target output result as the third information analysis result.

[0171] It can be seen that in the optional embodiment, a two-stage matching process is designed, which integrates the first information analysis result (context feature) and the second information analysis result (semantic feature), and combines multi-dimensional analysis (including matching degree calculation, applicability score and dynamic sorting) to realize step-by-step and accurate screening of knowledge items. The problem of response deviation caused by single matching dimension is solved, which is conducive to improving the accuracy and applicability of the target knowledge item obtained by screening; and for the two-stage matching result, a corresponding information filling and emotion adjustment mechanism is further set to realize intelligent adjustment between standardized knowledge items and personalized response content, solve the problems of low matching efficiency and mechanical response in traditional knowledge base application, improve the humanization degree of the output third information analysis result / overall intelligent customer system, thereby improving the user's interactive experience.

[0172] In yet another optional embodiment, the third information analysis submodule 3033 performs a two-stage matching operation on the first matching result according to the knowledge base to obtain a two-stage matching result corresponding to the first matching result, and the manner specifically includes:

[0173] For each knowledge item to be screened, a text similarity calculation operation is performed on the second information analysis result and the knowledge item to be screened according to the knowledge base to obtain a similarity calculation result between the second information analysis result and the knowledge item to be screened; the text similarity calculation operation includes a text preprocessing operation, a feature extraction operation and a similarity calculation operation;

[0174] A semantic annotation operation is performed on the second information analysis result and the knowledge item to be screened according to the knowledge base to obtain a semantic annotation result corresponding to the second information analysis result and the knowledge item to be screened; the semantic annotation operation includes a syntax structure analysis operation, a semantic role annotation operation and a semantic role comparison operation;

[0175] According to the similarity calculation result and the semantic annotation result corresponding to the knowledge item to be screened, a comprehensive similarity score between the second information analysis result and the knowledge item to be screened is calculated, and an applicability score of the knowledge item to be screened and the user's query demand and / or the conversation scene is calculated;

[0176] Taking the comprehensive similarity score and the applicability score corresponding to each knowledge item to be screened as a reference, a scoring and sorting operation is performed on all knowledge items to be screened, and any knowledge item to be screened within a preset ranking range is selected from the corresponding scoring and sorting result as a target knowledge item, and as a two-stage matching result.

[0177] In this optional embodiment, the text preprocessing operation includes word segmentation, stop word removal, stem extraction and word group reorganization; the text preprocessing is used to process the second information analysis result into a standardized text representation;

[0178] The semantic role labeling operation is used for labeling all semantic roles corresponding to the second information analysis result, and the semantic roles include an agent role, a patient role, a tool, time and location.

[0179] It can be seen that in the optional embodiment, by constructing the "text similarity-semantic role-two-dimensional score" corresponding three-level screening mechanism, the optimal response content is quickly located from the mass knowledge entries, the problems of semantic gap and low scene adaptation in the traditional knowledge matching are effectively solved, the service accuracy and adaptability of the B2B intelligent customer service system are improved, and the interactive experience and business processing efficiency of the user using the B2B intelligent customer service system are improved.

[0180] Embodiment four

[0181] Please refer to Figure 5 , Figure 5 is a structural schematic diagram of a B2B system intelligent customer service management and control device based on natural language processing disclosed by the embodiment of the application. As Figure 5 shown, the B2B system intelligent customer service management and control device based on natural language processing can include:

[0182] The memory 401 stores executable program codes.

[0183] The processor 402 is coupled with the memory 401.

[0184] The processor 402 invokes the executable program codes stored in the memory 401 to execute the steps in the B2B system intelligent customer service management and control method based on natural language processing described in the embodiment one or the embodiment two of the application.

[0185] Embodiment five

[0186] The embodiment of the application discloses a computer storage medium, which stores computer instructions, and when the computer instructions are invoked, the steps in the B2B system intelligent customer service management and control method based on natural language processing described in the embodiment one or the embodiment two of the application are executed.

[0187] Embodiment six

[0188] The embodiment of the application discloses a computer program product, which includes a non-transitory computer storage medium storing a computer program, and the computer program is operable to make a computer execute the steps in the B2B system intelligent customer service management and control method based on natural language processing described in the embodiment one or the embodiment two.

[0189] The system / device embodiments described above are only illustrative, wherein the modules illustrated as separate components can or can not be physically separated, and the components illustrated as modules can or can not be physical modules, i.e., can be located in one place or distributed to multiple network modules. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0190] Through the specific description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, and the computer software products can be stored in a computer storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disk storage, a magnetic disk storage, a magnetic tape storage, or any other computer readable medium that can be used to carry or store data.

[0191] Finally, it should be noted that: the B2B system intelligent customer service management and control method and system based on natural language processing disclosed by the embodiments of the application are only the preferred embodiments of the application, and are used to illustrate the technical solutions of the application, but not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that; the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.

Claims

1. A B2B system intelligent customer service management and control method based on natural language processing, characterized in that, The method includes: Obtain natural language query information from users through multiple channels, including query information triggered by users through a B2B platform; A preprocessing operation is performed on the natural language query information to obtain a preprocessing result for the natural language query information; the preprocessing is used to convert the natural language query information into recognizable text data; The preprocessing result is subjected to a target processing operation to obtain a target processing result for the preprocessing result; the target processing operation includes a first information analysis operation based on a context understanding module, a second information analysis operation based on a semantic analysis model, and a third information analysis operation based on a knowledge base; the target processing result is used to indicate the response item in response to the natural language query information; The step of performing target processing operations on the preprocessing results to obtain target processing results for the preprocessing results includes: According to the context understanding module, a first information analysis operation is performed on the preprocessing result to obtain a first information analysis result corresponding to the preprocessing result; the first information analysis operation includes at least one of the following operations: historical record query, session state analysis, and user preference analysis. Based on the semantic analysis model, a second information analysis operation is performed on the first information analysis result to obtain a second information analysis result corresponding to the first information analysis result; the second information analysis operation includes at least one of user query intent recognition, entity information parsing, and user sentiment analysis. Based on the knowledge base, a third information analysis operation is performed on the second information analysis result to obtain a third information analysis result corresponding to the second information analysis result, which is used as the target processing result; the third information analysis operation is used to match the response content in response to the second information analysis result; The context understanding module performs a first information analysis operation on the preprocessing result to obtain a first information analysis result corresponding to the preprocessing result, including: Based on a preset conversation state model, the conversation state after the user triggers the natural language query information is monitored in real time. The conversation state includes any one of the following states: conversation start, in progress, paused, or ended. When the session status is determined to be either the session start or the session in progress, it is detected whether there is a historical dialogue record corresponding to the preprocessing result. When the historical dialogue record is detected, information extraction is performed on the historical dialogue record according to the context understanding module to obtain a record summary corresponding to the historical dialogue record. The record summary includes at least one of the following: historical dialogue content with a correlation degree higher than a preset correlation threshold with the natural language query information, question type, and solution. The context understanding module performs user preference analysis on the historical dialogue records to obtain the user's query preferences, potential needs, and points of interest, which are used as user preference information. The record summary and the user preference information are correlated and fused with the preprocessing result to update the preprocessing result and obtain the first information analysis result; The step of performing a second information analysis operation on the first information analysis result according to the semantic analysis model to obtain a second information analysis result corresponding to the first information analysis result includes: Based on the semantic analysis model, user query intent recognition is performed on the first information analysis result to obtain the user's intent recognition result; the intent recognition result includes intent category and confidence score. Based on the semantic analysis model, entity information parsing is performed on the first information analysis result to obtain target entity information for the first information analysis result; the entity information parsing is used to extract keywords corresponding to the information query in the preprocessing result; the target entity information includes one or more entity keywords. Based on the semantic analysis model, sentiment analysis is performed on the first information analysis result to obtain the target sentiment and its corresponding sentiment intensity that triggered the natural language query information by the user; the target sentiment includes positive sentiment, negative sentiment, or neutral sentiment. The intent recognition result, the target entity information, and the target emotion and its corresponding emotion intensity are determined as the second information analysis result corresponding to the first information analysis result; The step of performing a third information analysis operation on the second information analysis result based on the knowledge base to obtain a third information analysis result corresponding to the second information analysis result includes: Based on the knowledge base, a first-level matching operation is performed on the second information analysis result to obtain a first-level matching result corresponding to the second information analysis result; the first-level matching operation includes knowledge item retrieval; the first-level matching result includes at least one knowledge item to be filtered. Based on the knowledge base, a second-level matching operation is performed on the first-level matching result to obtain a second-level matching result corresponding to the first-level matching result; the second-level matching operation includes in-depth analysis based on the first information analysis result and the second information analysis result, matching degree calculation, applicability scoring and ranking; the second-level matching result includes target knowledge items that meet the screening requirements; Based on the second information analysis result, information filling and sentiment adjustment are performed on the secondary matching result to obtain the target output result, which is used as the target processing result.

2. The intelligent customer service management method for B2B systems based on natural language processing according to claim 1, characterized in that, The step of performing a second-level matching operation on the first-level matching result based on the knowledge base to obtain a second-level matching result corresponding to the first-level matching result includes: For each knowledge item to be screened, a text similarity calculation operation is performed on the second information analysis result and the knowledge item to be screened according to the knowledge base to obtain the similarity calculation result between the second information analysis result and the knowledge item to be screened; the text similarity calculation operation includes text preprocessing operation, feature extraction operation and similarity calculation operation; Based on the knowledge base, semantic annotation operations are performed on the second information analysis result and the knowledge item to be screened to obtain the semantic annotation result corresponding to the second information analysis result and the knowledge item to be screened; the semantic annotation operation includes syntactic structure parsing operation, semantic role annotation operation, and semantic role comparison operation. Based on the similarity calculation result and the semantic annotation result corresponding to the knowledge item to be screened, calculate the comprehensive similarity score between the second information analysis result and the knowledge item to be screened, and calculate the applicability score between the knowledge item to be screened and the user's query needs and / or conversation scenarios; Based on the comprehensive similarity score and the applicability score corresponding to each of the knowledge items to be screened, all the knowledge items to be screened are ranked and sorted. Any knowledge item to be screened that ranks within a preset ranking range is selected from the corresponding ranking results and recorded as the target knowledge item, and used as the secondary matching result.

3. The intelligent customer service management method for B2B systems based on natural language processing according to claim 2, characterized in that, The text preprocessing operations include word segmentation, stop word removal, stemming, and phrase recombination; the text preprocessing is used to process the second information analysis result into a standardized text representation; The semantic role labeling operation is used to label all semantic roles corresponding to the second information analysis result. All semantic roles include agent role, patient role, tool, time and place.

4. A B2B intelligent customer service management system based on natural language processing, characterized in that, The system is used to execute the intelligent customer service management method for B2B systems based on natural language processing as described in any one of claims 1-3, and the system comprises: The acquisition module is used to acquire natural language query information fed back by users through multiple channels, including query information triggered by users through the B2B platform; A preprocessing module is used to perform preprocessing operations on the natural language query information to obtain a preprocessing result for the natural language query information; the preprocessing is used to convert the natural language query information into recognizable text data; The target processing module is used to perform target processing operations on the preprocessing results to obtain target processing results for the preprocessing results; the target processing operations include at least one of a first information analysis operation based on a context understanding module, a second information analysis operation based on a semantic analysis model, and a third information analysis operation based on a knowledge base; the target processing results are used to indicate response items in response to the natural language query information.

5. A B2B system intelligent customer service management device based on natural language processing, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the intelligent customer service management method for B2B systems based on natural language processing as described in any one of claims 1-4.

6. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked, are used to execute the intelligent customer service management method for a B2B system based on natural language processing as described in any one of claims 1-4.

Images