Customer service robot knowledge base management method and device

Abstract

The invention provides a customer service robot knowledge base management method and device. The method comprises the steps of obtaining multi-source data; performing problem semantic analysis on the missed problem data, and performing clustering processing on the missed problem data based on an analysis result to obtain an unsolved problem type set; on the basis of the unsolved question type set, the manual customer service answering data and the customer service robot knowledge base, constructing a knowledge association graph, and matching with the unsolved question type set on the basis of the knowledge association graph to obtain question candidate knowledge entries; for each knowledge entry in the question candidate knowledge entries, evaluating each knowledge entry based on user session log data, manual customer service answering data and a customer service robot knowledge base to obtain a confidence level of the knowledge entry; and updating each knowledge entry to the customer service robot knowledge base based on the confidence level of each knowledge entry. According to the method, the problem of knowledge base optimization delay under artificial interference is solved, and the knowledge base updating efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a management method and device for a customer service robot knowledge base. Background Art

[0002] With the continuous development of artificial intelligence (AI), intelligent customer service robots are increasingly being used in various fields. Leveraging natural language processing and machine learning algorithms, these robots enable intelligent question-and-answer interactions, providing users with more efficient and accurate service. The core of intelligent customer service robots is the knowledge base. Only by effectively managing and optimizing this knowledge base can high-quality Q&A services be ensured.

[0003] Currently, the knowledge base of customer service robots has certain limitations in practical application and urgently needs to be continuously expanded and improved during use. However, currently, knowledge base updates mainly rely on manual operations, that is, manual updates and optimization work is carried out after encountering new problems. This update and optimization model is cumbersome and consumes a lot of manpower and material resources. It can also easily lead to delays in knowledge base optimization, making it difficult to keep up with actual business needs. Summary of the Invention

[0004] The present invention provides a management method and device for a customer service robot knowledge base, which is used to solve the problem that the operation process of updating the optimization mode is cumbersome, which not only consumes a large amount of manpower and material resources, but also easily causes delays in knowledge base optimization under human interference. The knowledge base update efficiency is improved, and it can respond to business changes and user needs in a timely manner.

[0005] In a first aspect, the present invention provides a method for managing a customer service robot knowledge base, comprising:

[0006] Acquire multi-source data; the multi-source data includes user session log data, missed question data, and manual customer service answer data;

[0007] Performing problem semantic analysis on the missed problem data, and clustering the missed problem data based on the analysis result to obtain a set of unresolved problem types;

[0008] Based on the set of unresolved problem types, the manual customer service answer data, and the customer service robot knowledge base, a knowledge association graph is constructed, and candidate problem knowledge entries are obtained by matching the knowledge association graph with the set of unresolved problem types;

[0009] For each of the candidate knowledge items of the question, evaluate each knowledge item based on the user session log data, the manual customer service answer data, and the customer service robot knowledge base to obtain a confidence level of the knowledge item;

[0010] Each of the knowledge items is updated to the customer service robot knowledge base based on the confidence level of each of the knowledge items.

[0011] In a second aspect, the present invention further provides a management device for a customer service robot knowledge base, which is applied to the management method for a customer service robot knowledge base as described in the first aspect; the management device for the customer service robot knowledge base includes:

[0012] A multi-source data acquisition module is used to obtain multi-source data; the multi-source data includes user session log data, missed question data, and manual customer service answer data;

[0013] A semantic clustering analysis module, configured to perform semantic analysis on the missed question data, and cluster the missed question data based on the analysis results to obtain a set of unresolved question types;

[0014] A knowledge association graph construction module is used to construct a knowledge association graph based on the set of unresolved problem types, the manual customer service answer data, and the customer service robot knowledge base, and to match the knowledge association graph with the set of unresolved problem types to obtain candidate problem knowledge entries;

[0015] a confidence review module, configured to evaluate each knowledge item in the candidate knowledge items of the question based on the user session log data, the manual customer service answer data, and the customer service robot knowledge base to obtain a confidence level of the knowledge item;

[0016] An update management module is used to update each knowledge item to the customer service robot knowledge base based on the confidence level of each knowledge item.

[0017] In a third aspect, the present invention further provides an electronic device comprising: a memory for storing a computer software program; and a processor for reading and executing the computer software program, thereby implementing a method for managing a customer service robot knowledge base as described in any one of the above.

[0018] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, wherein the storage medium stores a computer software program, and when the computer software program is executed by a processor, the method for managing a customer service robot knowledge base as described in any one of the above is implemented.

[0019] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, which, when executed by a processor, implements a method for managing a customer service robot knowledge base as described in any one of the above.

[0020] The management method of the customer service robot knowledge base provided by the embodiment of the present invention can quickly and accurately find the set of unresolved problem types through semantic analysis and clustering of automatically acquired multi-source data, and then combine the matching of knowledge association graphs with unresolved problems to deeply understand the essence of user problems, automatically mine more accurate knowledge associations, reduce human and material resources, and provide accurate data support for subsequent updates and optimizations. Finally, the association evaluation of candidate knowledge items of the problem is carried out based on manual customer service answer data, user session log data and knowledge base content, which can ensure the accuracy and reliability of knowledge added to the knowledge base from multiple dimensions, and automatically update and optimize the knowledge base according to the evaluation confidence level without a lot of manual intervention, thereby solving the problem of knowledge base optimization delay under manual interference, improving the knowledge base update efficiency, and being able to respond to business changes and user needs in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 1 is a flow chart of a method for managing a customer service robot knowledge base provided by an embodiment of the present invention;

[0022] Figure 2 2 is a schematic diagram of the structure of a management device for a customer service robot knowledge base provided by an embodiment of the present invention;

[0023] Figure 3 An embodiment diagram of an electronic device provided by an embodiment of the present invention;

[0024] Figure 4 A diagram of an embodiment of a computer-readable storage medium provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0026] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the specified features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0027] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or explanation". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0028] See Figure 1 , Figure 1 : This is a flow chart of a method for managing a customer service robot knowledge base provided by the present invention. In an embodiment of the present invention, the method for managing a customer service robot knowledge base is executed by a knowledge base management device. Therefore, the method for managing a customer service robot knowledge base includes:

[0029] Step 10: Acquire multi-source data; the multi-source data includes user session log data, missed question data, and manual customer service answer data.

[0030] Optionally, the knowledge base management device connects to the user interaction terminal, the customer service robot server and the manual customer service workstation and establishes a data transmission channel, and uses API interfaces, data interface protocols (such as HTTP, WebSocket) and other methods to periodically (such as every second, every minute) obtain real-time transmission data. Among them, the user session log data covers the complete record of the user's interaction with the customer service robot, including the user ID, question time, question content, robot's reply content and user feedback information on the reply (such as follow-up questions, approval or denial, etc.); the missed question data refers to the user's question record data after the user asks the customer robot a question, and the customer robot cannot find a matching answer or cannot give an effective answer based on the existing customer robot knowledge base content. It is mainly used to accurately locate the knowledge blind spots of the current customer service robot knowledge base; the manual customer answer data is mainly extracted from the manual customer service system and comes from the records of manual customer service handling user problems. It contains information such as the time when the problem is transferred to the manual, the manual customer service answer text and the user confirmation status, so as to provide professional and effective problem solutions. The integration of source data can avoid the limitations of a single data source and more comprehensively cover the problem scenarios that users may encounter.

[0031] Furthermore, after obtaining user session log data, missed question data, and manual customer service answer data, the knowledge base management device performs preliminary cleaning on the original data to determine the accuracy and completeness of the data after obtaining the data, removing duplicate records, garbled characters, and irrelevant information. For example, for user session log data, if it is found that the same user repeatedly asks the same question in a short period of time, only one valid record will be retained.

[0032] Furthermore, in one embodiment, using an e-commerce customer service system as an example, user session logs are collected from the user's app or PC every five minutes. If a user asks, "Does this product support 7-day no-questions-asked returns?" and the customer service robot searches the customer service robot knowledge base and finds no matching answer, the question is automatically marked as a missed question. At the same time, if a human customer service representative intervenes and responds, "Except for customized products, all products can be returned within 7 days without any questions asked," the relevant answer is collected into the human customer service answer data.

[0033] Step 20: perform problem semantic analysis on the missed problem data, and cluster the missed problem data based on the analysis results to obtain a set of unresolved problem types.

[0034] Optionally, after the knowledge base management device obtains multi-source data, it processes the missed question data in the multi-source data through natural language processing technology, determines the lexical semantic features and semantic categories of each missed question in the missed question data, and performs cluster analysis processing based on the correlation between the lexical semantic features and the semantic categories to obtain a set of unresolved question types, as specifically described in steps 201 to 204.

[0035] Furthermore, the unresolved problem type set obtained by the knowledge base management device includes a set of all unresolved problem types. Again, taking the e-commerce customer service scenario as an example, if the collected unresolved questions include "Can I get a refund mid-delivery after receiving the goods?" and "Can I get a refund if I no longer want the goods?", these two questions are first subjected to natural language processing to obtain the corresponding lexical semantic features and semantic categories. Then, after clustering based on the correlation between the corresponding lexical semantic features and semantic categories, they are classified as unresolved problem types related to product refund policies.

[0036] Step 30: Based on the unresolved problem type set, manual customer service answer data and customer service robot knowledge base, a knowledge association graph is constructed, and the knowledge association graph is matched with the unresolved problem type set to obtain candidate problem knowledge entries.

[0037] Optionally, after determining the set of unresolved problem types, the knowledge base management device constructs a knowledge association graph based on the set of unresolved problem types, the manual customer service answer data, and the customer service robot knowledge base, using the problem attributes of each unresolved problem in the set of unresolved problem types as nodes and the relationships between the problem attributes as edges. This is described in detail in steps 3011 through 3015. This knowledge association graph allows for a quick and intuitive display of the relationships between unresolved problem types and various pieces of knowledge, enabling a rapid understanding of the structure and connections of the knowledge system.

[0038] Furthermore, after determining the knowledge association graph, the knowledge base management device first performs natural language processing on each unresolved problem in the unresolved problem type set. The semantic units obtained after natural language processing are mapped to nodes in the knowledge management graph, and the association paths are expanded and supplemented. Finally, each unresolved problem in the unresolved problem type set is matched, screened, and integrated with knowledge items to obtain subsequent knowledge items, as described in steps 3021 to 3025. This allows for the precise identification of candidate knowledge items related to the unresolved problem type, improving the accuracy and efficiency of knowledge matching.

[0039] Step 40 : For each knowledge item in the candidate knowledge items of the question, evaluate each knowledge item based on the user session log data, manual customer service answer data, and the customer service robot knowledge base to obtain a confidence level of the knowledge item.

[0040] Optionally, after the knowledge base management device determines the candidate knowledge items of the problem, it evaluates each knowledge item in the candidate knowledge items from three aspects: user session log data, manual customer service answer data, and customer service robot knowledge base, so as to obtain a confidence level for each knowledge item. Specifically, as described in steps 401 to 405, the confidence level of the knowledge item is evaluated by comprehensively considering multiple factors, which can comprehensively and objectively evaluate the quality and reliability of the knowledge item, avoid the one-sidedness of a single standard, and reduce the risk of erroneous knowledge entry.

[0041] Step 50: Update each knowledge item to the customer service robot knowledge base based on the confidence level of each knowledge item.

[0042] Optionally, after obtaining the confidence level for each knowledge item, the knowledge base management device adopts different update strategies according to the confidence level of each knowledge item, wherein the confidence level includes at least three levels, such as the first confidence level, the second confidence level and the third confidence level. The confidence standard of each confidence level can be adjusted according to the actual situation, as described in steps 501 to 503. It should be noted that the confidence level is not limited to three confidence levels, and can also be adjusted according to actual conditions and needs, thereby avoiding differences in the quality and reliability of knowledge items of different confidence levels. The use of a unified update method may lead to low-quality knowledge in the knowledge base or waste of audit resources. At the same time, the use of a hierarchical update strategy can perform targeted processing according to the actual situation of the knowledge items, which can reasonably allocate resources and improve the efficiency and quality of knowledge base updates.

[0043] The embodiment of the present invention can quickly and accurately find the set of unresolved problem types through semantic analysis and clustering of automatically acquired multi-source data. Combined with the matching of knowledge association graphs and unresolved problems, it can also deeply understand the essence of user problems and automatically mine more accurate knowledge associations, reducing human and material resources, and providing accurate data support for subsequent updates and optimizations. Finally, based on the artificial customer service answer data, user session log data and knowledge base content, the association evaluation of candidate knowledge items of the problem can ensure the accuracy and reliability of the knowledge added to the knowledge base from multiple dimensions, and automatically update and optimize the knowledge base according to the evaluation confidence level without a lot of manual intervention, which solves the problem of knowledge base optimization delay under manual interference, improves the knowledge base update efficiency, and can respond to business changes and user needs in a timely manner.

[0044] In one embodiment, steps 201 to 204 are described as follows:

[0045] Step 201 : performing word segmentation processing on the missed question data to obtain a plurality of vocabulary data, and performing feature extraction on the vocabulary data based on a preset semantic knowledge base to obtain vocabulary semantic features.

[0046] Optionally, the knowledge base management device uses a word segmentation tool (such as Jieba word segmentation, HanLP) to segment the missed question data. Take the missed question in the e-commerce customer service scenario: Where can I see the size chart of this dress as an example? After Jieba word segmentation, the vocabulary data such as this, clothes, of, size chart, where, and see are obtained. After the word segmentation is completed, when extracting features from the vocabulary data according to the preset semantic knowledge base, not only can we rely on word vector models (such as Word2Vec), but we can also use other semantic representation methods such as GloVe, FastText, etc. for analysis and processing. Taking the call of the Word2Vec word vector model in the preset semantic knowledge base as an example, the model has been pre-trained on a large-scale text corpus. Each word is input into the word vector model to generate the corresponding word vector. These word vectors are lexical semantic features. For example, the word "clothes" will obtain a vector of a specific dimension (such as 300 dimensions) under the Word2Vec model. This vector contains the semantic information of clothes in the semantic space.

[0047] Step 202 : performing syntactic structure analysis on the lexical semantic features to obtain syntactic structure information, and performing semantic role labeling on the lexical semantic features based on the syntactic structure information to obtain semantic role information.

[0048] Optionally, after determining the lexical data and lexical semantic features of the segmented missing question data, the knowledge base management device performs a syntactic structure analysis on the lexical semantic features. This not only determines basic grammatical relationships based on the dependency syntactic analyzer, but also introduces semantic dependency analysis to further explore the semantic dependency relationships between words in the sentence, such as causal relationships, purpose relationships, etc. It should be noted that for some long and difficult sentences, a recursive approach can be used to perform syntactic structure analysis, breaking the sentence into multiple clauses, analyzing the syntactic structure of each clause separately, and then synthesizing the syntactic structures of all clauses to obtain syntactic structure information for the entire sentence.

[0049] Furthermore, after the knowledge base management device determines the syntactic structure information, it performs semantic role annotation on the lexical semantic features according to the syntactic structure information. It can not only perform annotation of common semantic roles (implementer, recipient, etc.), but also perform customized semantic role annotation according to specific business fields and problem characteristics. For example, in the e-commerce customer service scenario, product attribute roles (used to describe the color, size and other attributes of the product) and transaction behavior roles (such as purchase, return, etc.) can be added. Moreover, for some situations where semantic roles are ambiguous, more accurate annotation can be performed in combination with the semantic constraint rules in the semantic knowledge base. For example, when a user buys a piece of clothing, the clothing is the recipient and also belongs to the product attribute role. Through the semantic constraint rules, it can be made clear that in the context of the sentence, the focus is more on the product attribute role, and then semantic role annotation is performed to determine the final semantic role information.

[0050] Furthermore, syntactic structure analysis clarifies the grammatical relationships between words, helping to understand the semantic combination logic of sentences. Semantic role annotation further refines the role of components at the semantic level, making text semantic understanding more in-depth and comprehensive, and providing richer and more accurate information for semantic reasoning.

[0051] Step 203 : performing semantic reasoning on the lexical semantic features, syntactic structure information, and semantic role information based on a preset semantic reasoning rule base to determine the semantic category.

[0052] Optionally, when pre-building a preset semantic reasoning rule base, the knowledge base management device may include not only simple rule matching based on lexical semantic features, syntactic structure, and semantic role information, but also inference rules based on probabilistic statistics. For example, by statistically analyzing a large amount of historical miss question data, the probability of belonging to a certain semantic category under certain combinations of lexical semantic features, syntactic structure, and semantic role is calculated. For example, in historical data, when the syntactic structure of "product name + failure + function description" appears, with the agent being the user and the object being the product, there is an 80% probability that the data belongs to the product failure feedback semantic category. When performing semantic reasoning, combined with probability information, when encountering miss question data that matches this pattern, not only rule matching is performed, but also the probability of belonging to the product failure feedback semantic category is further determined based on probability, thereby improving the accuracy of semantic reasoning.

[0053] Step 204 , clustering the missed question data using lexical semantic features and semantic categories to obtain a set of unresolved question types.

[0054] Optionally, the knowledge base management device analyzes and clusters the missed question data based on the obtained lexical semantic features and semantic categories, and the association between the lexical semantic features and semantic categories, to obtain a set of unresolved question types, as specifically described in steps 2041 to 2045.

[0055] The embodiment of the present invention deeply analyzes the text semantics from multiple levels through multiple steps such as lexical semantic feature extraction, syntactic structure analysis, semantic role labeling and semantic reasoning, avoiding the one-sidedness of simple text processing, making semantic understanding more accurate and comprehensive, and providing a reliable semantic foundation for knowledge base management.

[0056] In one embodiment, steps 2041 to 2044 are described as follows:

[0057] Step 2041: construct a semantic hierarchy tree based on the association between semantic categories and lexical semantic features.

[0058] Optionally, the knowledge base management device sorts out the acquired semantic categories and lexical semantic features, first determining the top-level general semantic categories, such as product-related issues, and then gradually subdividing them according to the inclusion and subordination relationships between the semantic categories. For example, product-related issues can be divided into subcategories such as electronic product issues and daily necessities issues. For electronic product issues, they can be further subdivided into mobile phone issues, computer issues, etc. In terms of lexical semantic features, if lithium batteries belong to the battery category, and batteries are related to mobile phone issues (such as mobile phone battery-related issues), a semantic hierarchy tree is constructed according to this hierarchical relationship. A graph data structure is used to store the semantic hierarchy tree, where each node represents a semantic category or a vocabulary set with specific semantics, and the edges between nodes represent semantic inclusion or subordination relationships.

[0059] Furthermore, in one embodiment, using an e-commerce customer service scenario as an example, semantic categories include clothing issues and digital product issues. Digital product issues further include subcategories such as camera issues and mobile phone issues. Among the lexical semantic features, lens belongs to the camera category, and lithium battery belongs to the mobile phone battery category. In the constructed semantic hierarchy tree, digital product issues is the upper node, camera issues and mobile phone issues are its lower child nodes, and lexical semantic features such as lens and lithium battery are assigned to the corresponding lower nodes.

[0060] Step 2042: locate the unresolved issues in the missed issue data and the nodes in the semantic hierarchy tree to obtain unresolved issue nodes.

[0061] Optionally, the knowledge base management device locates each unresolved problem in the missed problem data in the semantic hierarchy tree according to its semantic analysis results (including semantic categories and lexical semantic features). First, determine the semantic category to which the unresolved problem belongs, and find the corresponding top-level or middle-level category node in the semantic hierarchy tree. Then, based on the specific lexical semantic features involved in the unresolved problem, further search for more precise sub-nodes under the category node. For example, if the unresolved problem is what to do if the lithium battery of a mobile phone has short battery life, its semantic category belongs to mobile phone problems, and the lexical semantic features point to lithium batteries. In the semantic hierarchy tree, first locate the mobile phone problem node, and then further locate the lithium battery problem node under the mobile phone battery problem. This node is the unresolved problem node.

[0062] Step 2043 , performing a path search on the unresolved problem nodes in the semantic hierarchy tree to obtain the unresolved problem node path, and performing in-depth processing on any two unresolved problem nodes based on a preset hierarchical correlation function to obtain a hierarchical correlation degree.

[0063] Optionally, the knowledge base management device searches for each unsolved problem node starting from the root node and records all the nodes passed through to form an unsolved problem node path. The process of performing deep processing on any two unsolved problem nodes according to a preset hierarchical correlation function is as follows: Let the unsolved problem node path of problem A be P A =(n A1 ,n A2 ,…,n Am ), the unsolved problem node path of problem B is P B =(n B1 ,n B2 ,…,n Bn ). The preset hierarchical correlation function is Where k represents starting from the root node, P A and P B The maximum depth of the same node on the path, m and n are P A and P B The total length of the path. For example, the path of problem A is electronic product problem - mobile phone problem - mobile phone battery problem - lithium battery problem, with a depth of 4; the path of problem B is electronic product problem - mobile phone problem - mobile phone charging problem, with a depth of 3. Problems A and B are the same before the mobile phone problem node, and the depth of the mobile phone problem node is 2, so k = 2, m = 4, n = 3, That is, the hierarchical correlation between question A and question B is The hierarchical association metric quantifies the degree of association of unsolved problems in the semantic hierarchy, taking into account the hierarchical relationship of semantics. Compared with simple distance calculation, it is more in line with semantic logic, provides a reasonable similarity metric for subsequent clustering, and helps to accurately cluster semantically related problems together.

[0064] Step 2044 , taking each unresolved problem in the missed problem data as an initial cluster, the initial clusters are merged based on the hierarchical association degree to obtain an unresolved problem type set.

[0065] Optionally, the knowledge base management device initially regards each unsolved problem in the missed problem data as a separate cluster (ie, initial cluster). Then, for all cluster pairs (C i ,C j ), calculate the average hierarchical correlation between them Where R(A,B) is the hierarchical correlation between question A and question B, |C i | and |C j Cluster C i and C j and select The two largest clusters are merged. This process is repeated until a stopping condition is met, such as the number of clusters reaching a preset value K, or the average hierarchical association of the clusters no longer increases significantly. The resulting K clusters are the set of unresolved problem types.

[0066] Furthermore, in one embodiment, it is assumed that there are three initial clusters C1, C2, and C3, which contain questions A1, A2; B1, B2; C1, C2 respectively. Calculate the average hierarchical correlation between C1 and C2 The average hierarchical correlation between C1 and C3 The average hierarchical correlation between C2 and C3 .like If the maximum value is reached, C1 and C2 are merged into a new cluster, and the iterative calculation and merging are continued until the stopping condition is met.

[0067] This embodiment of the present invention begins by constructing a semantic hierarchy tree, systematically sorting out the hierarchical relationships between semantic categories and lexical semantic features. It then locates, analyzes, and clusters missed question data based on the semantic hierarchy, achieving a comprehensive and in-depth understanding of question semantics while avoiding fragmented processing. By calculating hierarchical relevance and merging clusters based on this, operations are closely centered around the semantic hierarchy, conforming to the inherent logic of semantics. Compared to traditional clustering methods, this method can more accurately reflect the semantic connections between questions, making the clustering results more logical and reasonable.

[0068] In one embodiment, steps 3011 to 3015 are described as follows:

[0069] Step 3011: perform attribute extraction on the unresolved issues in the unresolved issue type set to obtain a problem attribute set.

[0070] Optionally, the knowledge base management device first uses a part-of-speech tagging tool to determine the keyword words such as nouns and adjectives for each unresolved problem in the problem type set, and then uses a named entity recognition model to identify the entities and attribute-related words involved in the problem. For example, for the question: Which brand of laptop has good heat dissipation, the brand, laptop, heat dissipation and other keywords are determined through part-of-speech tagging, and then the attributes of the laptop (entity) such as brand and heat dissipation performance are extracted through the named entity recognition model, and these extracted attributes are summarized to form a problem attribute set. By accurately extracting problem attributes, the foundation is laid for the subsequent knowledge association and graph construction around the problem attributes, and the direction of the association between the problem and the knowledge base knowledge is clarified, making the knowledge association more targeted.

[0071] Step 3012: classify the problem attribute set to obtain problem attribute categories, and construct a hierarchical structure between the problem attributes based on the hierarchical associations between the problem attributes in the problem attribute categories.

[0072] Optionally, the knowledge base management device categorizes the extracted question attributes into basic attributes, key attributes, and derived attributes based on pre-defined attribute classification criteria. Basic attributes are typically represented as inherent attributes directly related to the question object, such as brand and model; key attributes are represented as attributes that play a key role in answering the question, such as heat dissipation performance for laptop computer heat dissipation issues; and derived attributes are represented as more detailed attributes associated with the key attributes, such as heat dissipation design and heat dissipation materials, which are derived attributes of heat dissipation performance. A hierarchical structure is then constructed using a tree-like data structure based on the semantic associations and hierarchical relationships between the attributes, with the root node being the question object, the child nodes being attributes of different categories, and the hierarchical relationships between the attributes.

[0073] Step 3013: Based on the hierarchical structure between question attribute categories and question attributes, matching is performed with the comprehensive knowledge in the customer service robot knowledge base and manual customer service answer data to obtain question attribute knowledge entries.

[0074] Optionally, the knowledge base management device traverses each attribute in the problem attribute category and hierarchy, and performs retrieval and matching in the customer service robot knowledge base and manual customer answer data. Keyword matching, semantic matching and other methods can be used, wherein keyword matching is used to directly search for knowledge items containing attribute keywords; semantic matching is used to use semantic similarity calculation (such as cosine similarity) to match the semantics of the attribute and the semantics of the knowledge items in the knowledge base. For example, for the heat dissipation performance attribute of a laptop computer, in the computer-related knowledge base, keyword matching is used to search for knowledge items containing heat dissipation performance, and semantic matching is used to search for knowledge items with similar semantics to heat dissipation performance (such as heat dissipation effect). The screened-out knowledge items are the problem attribute knowledge items. By matching the attributes with the knowledge base and manual customer service answer data, it is possible to accurately screen out knowledge items related to the problem, and associate the knowledge in the knowledge base according to the problem attributes.

[0075] Step 3014: Analyze the relationship between question attributes in the question attribute knowledge items and construct a question attribute relationship model.

[0076] Optionally, the knowledge base management device conducts in-depth analysis based on the acquired problem attribute knowledge items, and uses text mining and relationship extraction technology (such as a relationship extraction model based on deep learning) to identify logical relationships such as causality, association, and influence between attributes. For example, in the knowledge item about laptop heat dissipation, it is found that the use of copper tube heat dissipation design will improve heat dissipation performance, and a positive influence relationship between heat dissipation design and heat dissipation performance is established. These identified attribute relationships are integrated to construct a problem attribute relationship model, which is represented by a graph structure. The nodes are problem attributes, and the edges are the relationships between attributes. The weight of the edges can be set according to the strength of the relationship (such as the degree of certainty of the causal relationship).

[0077] Step 3015: Based on the question attribute relationship model, a knowledge association graph is constructed with question attributes as nodes and relationships between question attributes as edges.

[0078] Optionally, the knowledge base management device is based on the problem attribute relationship model and uses a graph database (such as Neo4j) or a graph computing framework (such as GraphX) to construct a knowledge association graph. The problem attributes are used as nodes in the graph, and each node contains information such as the attribute name and attribute description; the relationship between the attributes is used as an edge, and the edge type (such as causality, association) and weight are set according to the problem attribute relationship model. At the same time, based on the relationship between the attributes, relevant knowledge is further expanded from the knowledge base to enrich the content of the graph. For example, from the heat dissipation material node to the manufacturer, characteristics and other knowledge of the material, the corresponding nodes and edges are added to the graph. By constructing a knowledge association graph with problem attributes as the core, the association relationship between problem-related knowledge is intuitively presented, which is highly targeted and practical. By continuously enriching the details of the graph through attribute expansion, it can better support the customer service robot to perform knowledge retrieval and problem answering based on the graph, thereby improving the quality and efficiency of customer service.

[0079] The embodiment of the present invention starts from the extraction of problem attributes, gradually constructs a knowledge association map, and performs knowledge matching and association around the attributes of the problem itself. It can closely fit the problem requirements, accurately associate relevant knowledge in the knowledge base, avoid interference from irrelevant knowledge, and clearly display the logical deduction relationship between knowledge through attribute classification, hierarchical structure construction and attribute relationship model construction, so that the knowledge association map is not only a knowledge collection, but also a knowledge network with logical connotations, which is convenient for users to understand and apply knowledge.

[0080] In one embodiment, steps 3021 to 3025 are described as follows:

[0081] Step 3021 , performing semantic structure decomposition on the unresolved issues in the unresolved issue type set to obtain multiple semantic units.

[0082] Optionally, the knowledge base management device uses syntactic analysis and semantic analysis tools in natural language processing technology (such as Stanford syntactic analyzer and semantic role labeling tool) to analyze each unresolved problem in the unresolved problem type set. First, the grammatical structure of the sentence is determined through syntactic analysis, such as subject, predicate, object, attributive, adverbial, complement and other components, and then the sentence is split into multiple semantic units according to semantic function based on semantic analysis. For example, for the unresolved problem: how to optimize the voice wake-up function of the smart speaker, it is clarified through syntactic analysis that how to optimize is the manner adverbial, the smart speaker is the subject, and the voice wake-up function is the object. Semantic analysis further decomposes semantic units such as smart speakers, voice wake-up functions, and optimization methods. Refining unresolved problems into semantic units provides a basis for subsequent precise matching in the knowledge association graph.

[0083] Step 3022: Map the semantic unit to the node in the knowledge association graph to obtain a semantic unit mapping node.

[0084] Optionally, the knowledge base management device traverses the nodes in the knowledge association graph and compares each semantic unit with the graph node using a semantic similarity calculation method (such as cosine similarity, edit distance, etc.). Based on the text content and semantic features of the semantic unit, search for nodes that match it. If there is a node in the knowledge association graph that is semantically similar or identical to the semantic unit, then the node is used as the mapping node of the semantic unit. For example, for the semantic unit smart speaker, search for nodes containing smart speaker keywords or having relevant semantic descriptions in the knowledge association graph. If a smart speaker product category node is found, then the node is the mapping node of the smart speaker semantic unit.

[0085] Step 3023, starting from the semantic unit mapping node, expands the association path in the knowledge association graph to obtain an initial association path, and supplements the initial association path based on the customer service robot knowledge base and external knowledge sources to obtain a target association path.

[0086] Optionally, the knowledge base management device starts from the semantic unit mapping node and performs a path search in the knowledge association graph based on the connection relationship between nodes (such as attribute association, causal association, etc.). A depth-first search or breadth-first search algorithm can be used to explore other nodes associated with the mapping node to form an initial association path. For example, starting from the smart speaker mapping node, relevant nodes such as the smart speaker hardware configuration and the smart speaker software algorithm are found through the association relationship to form an initial association path. Then, the customer service robot knowledge base and external knowledge sources (such as industry authoritative websites, professional forums, etc.) are queried to obtain supplementary knowledge related to the initial association path, and the initial association path is improved to obtain the target association path. For example, knowledge about the optimization of smart speaker microphone arrays is obtained from a professional audio technology website and added to the association path related to the voice wake-up function.

[0087] Step 3024 , for each unresolved problem in the unresolved problem type set, matching and screening are performed based on the target association path and the unresolved problem to obtain a knowledge entry.

[0088] Optionally, for each unresolved problem, the knowledge base management device filters out closely related knowledge items on the target association path based on its semantic features and problem requirements. The screening can be performed by calculating the semantic similarity, relevance score, etc. of the knowledge items on the target association path. For example, for the unresolved problem: how to optimize the voice wake-up function of the smart speaker, the semantic similarity of each knowledge item is calculated on the target association path, and a corresponding semantic similarity threshold is set. For example, if the semantic similarity between the smart speaker microphone array optimization method and the problem is greater than or equal to the semantic similarity threshold, and the semantic similarity between the smart speaker appearance design features is less than the semantic similarity threshold, then the knowledge items greater than or equal to the semantic similarity threshold are filtered out as knowledge items related to the unresolved problem. It should be noted that there can be multiple knowledge items corresponding to each unresolved problem.

[0089] Step 3025: Merge each unresolved problem in the unresolved problem type set and its corresponding knowledge entry to obtain a candidate problem knowledge entry.

[0090] Optionally, the knowledge base management device integrates each unresolved problem with the knowledge items it has screened out to form a candidate problem knowledge item. A data structure (such as a dictionary, a list, etc.) can be used to store this correspondence, and each candidate problem knowledge item contains the text content of the unresolved problem and a set of knowledge items that match it. For example, for the unresolved problem: how to optimize the voice wake-up function of the smart speaker and its screened out smart speaker microphone array optimization method, smart speaker voice wake-up algorithm improvement strategy and other knowledge items, are integrated into a candidate problem knowledge item and recorded as: {How to optimize the voice wake-up function of the smart speaker: [smart speaker microphone array optimization method, smart speaker voice wake-up algorithm improvement strategy]}.

[0091] The embodiment of the present invention starts with semantic structure decomposition, and gradually matches the semantics and attributes of the problem with the knowledge association graph through operations such as mapping, association path expansion and screening. It can accurately locate knowledge related to unsolved problems, improve the accuracy of knowledge matching, and avoid interference from irrelevant knowledge; and through association path expansion and supplementation in combination with internal and external knowledge sources, it can mine potential knowledge in the knowledge association graph and external related knowledge, making the acquired knowledge items more comprehensive and providing rich knowledge support for customer service robots to answer complex questions.

[0092] In one embodiment, steps 401 to 405 are described as follows:

[0093] Step 401: Search and match paths related to knowledge items in the knowledge association graph to obtain knowledge item path information.

[0094] Optionally, the knowledge base management device uses the query function of the graph database (such as Neo4j's Cypher query language) to perform a path search in the knowledge association graph with the knowledge item as the target node. Starting from other nodes in the knowledge association graph, find the path connected to the knowledge item. For example, for the knowledge item "Check whether the printer cartridge is out of ink", the path from the node of the problem "The printer cannot print" to the node of the knowledge item is searched in the knowledge association graph. The path information such as "Printer cannot print - Printer hardware check - Cartridge check - Check whether the printer cartridge is out of ink" may be obtained, and the node and edge information on the path is recorded.

[0095] Step 402: Search and match the information related to the knowledge item in the manual customer service answer data to obtain the knowledge item context information.

[0096] Optionally, the knowledge base management device uses information retrieval technology (such as inverted index, keyword matching, etc.) to search for content related to knowledge items in the manual customer service answer data. The manual customer service answer data is preprocessed, such as word segmentation, removal of stop words, etc. Then, search and match are performed in the data based on the keywords of the knowledge items. For example, for the knowledge item to check whether the printer cartridge is out of ink, the manual customer service answer data is searched for conversation records containing keywords such as printer cartridge and out of ink, and context information such as descriptions of printer failures, previously given solutions, and user feedback in these conversation records is obtained. Obtaining knowledge item context information from manual customer service answer data can reflect the application scenarios and user feedback of knowledge items in actual customer service conversations.

[0097] Step 403 : Perform knowledge logic analysis on the knowledge item path information, and match the logic analysis result with the knowledge item context information based on a matching function to obtain a first degree of fit.

[0098] Optionally, the knowledge base management device analyzes the knowledge item path information and sorts out the knowledge logical relationship therein, such as cause-effect relationship, step relationship, etc. For example, for the printer cannot print - printer hardware check - ink cartridge check - check whether the printer ink cartridge is out of ink, its logic is analyzed to be gradually deduced from the printer failure phenomenon to the possible hardware cause (ink cartridge problem). Then, according to the pre-built matching function (such as a function based on semantic similarity), the semantic representation of the knowledge item path information is set as S path , the semantic representation of the context information of the knowledge item is S context , which matches the matching function f(S path ,S context), and the first degree of fit is obtained by calculating the semantic similarity between the two. For example, using the cosine similarity formula The cosine value of the semantic vectors of the two is calculated as the first fit score. Through knowledge logic analysis and fit matching, we can preliminarily determine whether the logic of the knowledge items in the knowledge association graph is consistent with their application in actual customer service conversations.

[0099] Step 404 : Perform sentiment analysis on the user conversation data related to the knowledge item in the user conversation log data, and perform a preliminary optimization on the first fit based on the sentiment analysis to obtain a second fit.

[0100] Optionally, the knowledge base management device uses sentiment analysis technology (such as a sentiment analysis model based on deep learning, such as BERT-for-Sentiment-Analysis) to analyze the user conversation content related to the knowledge items in the user conversation log data. The user conversation data is input into the sentiment analysis model, and the user's emotional tendency score is output. The range can be set between [-1,1], where -1 represents extremely negative, 1 represents extremely positive, and 0 represents neutral. For example, if the user mentioned in the conversation: After following this method, the printer still cannot print, which is very disappointing, the emotional tendency score obtained by sentiment analysis is -0.6. The first degree of fit is adjusted according to the emotional tendency score. If the emotional tendency is positive, the first degree of fit score is increased; if it is negative, the score is reduced to obtain the second degree of fit. For example, if the first fit is 0.6 and the sentiment score is -0.6, the second fit can be calculated by adjusting the formula (e.g., second fit = first fit + sentiment score × adjustment coefficient, where the adjustment coefficient can be set to 0.2 based on experience) to 0.6 + (-0.6) × 0.2 = 0.48. Optimizing fit by factoring in user sentiment can more comprehensively reflect the effectiveness of knowledge items in actual user experience, avoiding the one-sidedness of evaluating only based on knowledge logic and contextual information, and making the evaluation results more relevant to actual application scenarios.

[0101] Step 405: Perform an effect analysis on the historical update feedback data related to the knowledge items in the customer service robot knowledge base, and perform a secondary optimization on the second fit based on the effect analysis results to obtain the target fit, and determine the confidence level corresponding to each knowledge item based on the target fit.

[0102] Optionally, the knowledge base management device queries the customer service robot's knowledge base for historical update feedback data related to knowledge items, such as the success rate of problem solving after knowledge item updates, the frequency of users asking related questions again, and so on. This data is analyzed to evaluate the historical update effect of the knowledge item. For example, if the success rate of problem solving for a knowledge item before the update was 30% and increased to 50% after the update, it indicates that the update effect is good. The second fit is then optimized based on the historical update effect. If the update effect is good, the second fit is improved using the formula (e.g., target fit = second fit × (1 + improvement coefficient), where the improvement coefficient is set based on the improvement effect, here set to 0.1). If the update effect is poor, the second fit is reduced to obtain the final target fit. The confidence level of the knowledge item is then determined based on pre-set confidence level classification criteria (e.g., a target fit greater than 0.8 is a high confidence level (i.e., the first confidence level), 0.5-0.8 is a medium confidence level (i.e., the second confidence level), and less than 0.5 is a low confidence level (i.e., the third confidence level)). By combining the historical knowledge update effect to conduct secondary optimization of the fit, further improving the evaluation from the perspective of the long-term application and optimization effect of the knowledge items, and comprehensively considering multiple factors to determine the confidence level, the evaluation results of the knowledge items are made more accurate and comprehensive, thereby improving the quality assessment level of the knowledge items in the knowledge base.

[0103] This embodiment of the present invention integrates multiple data points, including knowledge association graph path information, contextual information about manual customer service responses, user sentiment, and historical update feedback, to conduct an evaluation. This avoids the limitations of evaluation based on a single information source and comprehensively covers the performance of knowledge items in terms of their knowledge system, practical applications, user experience, and long-term effects. Furthermore, by considering user sentiment and historical update effects, and closely integrating feedback from the application of knowledge items in actual customer service scenarios, the evaluation results are more closely aligned with the actual usage of the knowledge items, more accurately reflecting their effectiveness and reliability in real-world applications.

[0104] In one embodiment, steps 501 to 503 are described as follows:

[0105] Step 501: If the confidence level corresponding to each knowledge item is the first confidence level, the knowledge item is directly updated to the customer service robot knowledge base.

[0106] Optionally, when the knowledge base management device detects that the confidence level obtained by each knowledge item after step 405 is the first confidence level (high confidence level, target fit is greater than or equal to 0.8), an update operation is automatically triggered. The knowledge item is added to the corresponding category directory in the customer service robot knowledge base, and the index of the knowledge base is updated for subsequent quick retrieval. For example, in the e-commerce customer service knowledge base, the knowledge item "A certain brand of sports shoes uses a breathable mesh material and is comfortable to wear" has a confidence level of 0.85, which belongs to the first confidence level. The system automatically updates it to the knowledge base under the sports shoe product introduction category. By directly updating high-confidence knowledge items, the knowledge base update efficiency is improved, and rigorously evaluated and reliable knowledge can be quickly added to the knowledge base, so that the customer service robot has the ability to answer relevant questions in a timely manner, meeting the user's needs for quick acquisition of accurate knowledge.

[0107] Step 502: If the confidence level corresponding to each knowledge item is the second confidence level, the knowledge item is updated to the customer service robot knowledge base after manual review and confirmation.

[0108] Optionally, when the knowledge base management device detects that the confidence level obtained by each knowledge entry after step 405 is the second confidence level (medium confidence level, target fit is between 0.5-0.8), the knowledge entry is automatically pushed to the manual review interface. Manual reviewers (such as customer service supervisors, domain experts, etc.) review the accuracy, completeness, applicability, etc. of the knowledge entries. During the review process, the manual reviewer will refer to the logic of the knowledge entry in the knowledge association map, the application status in the user session log, historical update feedback and other information. Determine whether the knowledge entry is accurate and reasonable. If the review is passed, the manual reviewer will confirm and update the knowledge entry to the customer service robot knowledge base; if not, proceed to step 503.

[0109] Step 503: If the confidence level corresponding to each knowledge item is the third confidence level, the feature information of the knowledge item is extracted, data mining is performed in multiple data sources based on the extracted feature information, and the mined data is evaluated for relevance. Based on the relevance evaluation results and the preset standard threshold, the data that meets the requirements is screened out, and then the screened multi-source data is fused. The knowledge item is updated and improved based on the fused data, and a second confidence evaluation is performed on the updated and improved knowledge item.

[0110] Optionally, when the knowledge base management device detects that the confidence level of each knowledge item after step 405 is the third confidence level (medium confidence level, target fit less than 0.5), and receives a knowledge item that failed the review in step 502, it uses text analysis technology to analyze the knowledge item and extract characteristic information such as core concepts, key technologies, and main viewpoints. For example, for the knowledge item "a new environmentally friendly material can be used for building exterior walls", characteristic information such as the new environmentally friendly material and building exterior walls is extracted.

[0111] Next, based on the extracted characteristic information, data mining is conducted across multiple data sources (such as professional academic databases, industry reports, and authoritative websites). Tools such as web crawlers and database query interfaces are used to obtain relevant data from these diverse sources. For example, for characteristic information on new environmentally friendly materials, relevant research papers are searched in academic databases, performance indicators and application cases are searched in industry reports, and market trends and the latest information are collected from authoritative websites.

[0112] Assume that the amount of data related to knowledge items mined from the pth data source is D p , the total number of data sources is s, and the relevance of the mined data is evaluated to determine the relevance between the data and the knowledge items. Using the formula where NumD p Represented as the data D mined from the pth data source p The number of elements associated with the knowledge item; TotalnumD p Represented as the data D mined from the pth data source p The total number of elements in the data source. Weight (p) represents a weight coefficient set based on the reliability of the data source. For example, academic databases have a higher weight of 0.8, industry reports have a weight of 0.6, and authoritative websites have a weight of 0.4. This formula is used to calculate the relevance score of each data source and select data with high relevance.

[0113] The selected multi-source data is fused to remove duplicate information and integrate complementary information to form supporting evidence for the knowledge item. Based on the fused data, the knowledge item is updated and improved, and the confidence level is reassessed, i.e., a secondary confidence assessment is performed on the updated and improved knowledge item. For knowledge items with low confidence levels, a series of multi-source data mining and fusion operations are performed to fully utilize information from multiple channels to improve their accuracy and reliability. A secondary confidence assessment is then performed to ensure the quality of the knowledge ultimately stored. This, in turn, broadens the knowledge source channels of the knowledge base, enriches the knowledge content, and further enhances the overall practicality and credibility of the knowledge base.

[0114] The embodiment of the present invention adopts different processing methods according to the different confidence levels of knowledge items. Direct update, manual review, multi-source data mining and improvement processes are designed for high, medium and low confidence levels respectively, making the processing more targeted. It not only ensures the rapid storage of high-confidence level knowledge, but also reasonably optimizes medium and low confidence level knowledge, thereby improving the overall efficiency and quality of knowledge base updates.

[0115] Furthermore, the management device of the customer service robot knowledge base provided by the present invention is described below. The management device of the customer service robot knowledge base described below and the management method of the customer service robot knowledge base described above can refer to each other.

[0116] Optional, see Figure 2 , Figure 2 It is a structural diagram of the management device of the customer service robot knowledge base provided by the present invention, and the management device of the customer service robot knowledge base includes.

[0117] Multi-source data acquisition module 210, used to obtain multi-source data; multi-source data includes user session log data, missed question data and manual customer service answer data;

[0118] Semantic cluster analysis 220 is used to perform semantic analysis on the missed question data and cluster the missed question data based on the analysis results to obtain a set of unresolved question types;

[0119] The knowledge association graph construction module 230 is used to construct a knowledge association graph based on the unresolved problem type set, manual customer service answer data, and the customer service robot knowledge base, and to match the knowledge association graph with the unresolved problem type set to obtain candidate problem knowledge entries;

[0120] A confidence review module 240 is configured to evaluate each candidate knowledge item in the question based on user session log data, manual customer service answer data, and the customer service robot knowledge base to obtain a confidence level for the knowledge item;

[0121] The update management module 250 is used to update each knowledge item to the customer service robot knowledge base based on the confidence level of each knowledge item.

[0122] The embodiment of the present invention can quickly and accurately find the set of unresolved problem types through semantic analysis and clustering of automatically acquired multi-source data. Combined with the matching of knowledge association graphs and unresolved problems, it can also deeply understand the essence of user problems and automatically mine more accurate knowledge associations, reducing human and material resources, and providing accurate data support for subsequent updates and optimizations. Finally, based on the artificial customer service answer data, user session log data and knowledge base content, the association evaluation of candidate knowledge items of the problem can ensure the accuracy and reliability of the knowledge added to the knowledge base from multiple dimensions, and automatically update and optimize the knowledge base according to the evaluation confidence level without a lot of manual intervention, which solves the problem of knowledge base optimization delay under manual interference, improves the knowledge base update efficiency, and can respond to business changes and user needs in a timely manner.

[0123] See also Figure 3 , Figure 3 This is a diagram of an embodiment of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, an embodiment of the present invention provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, the following steps are implemented:

[0124] Acquire multi-source data; multi-source data includes user session log data, missed question data, and manual customer service answer data;

[0125] Perform problem semantic analysis on the missed problem data, and cluster the missed problem data based on the analysis results to obtain a set of unresolved problem types;

[0126] Based on the set of unresolved problem types, manual customer service answer data, and the customer service robot knowledge base, a knowledge association graph is constructed. The knowledge association graph is then matched with the set of unresolved problem types to obtain candidate knowledge entries for the problem.

[0127] For each candidate knowledge item in the question, each knowledge item is evaluated based on user session log data, manual customer service answer data, and the customer service robot knowledge base to obtain the confidence level of the knowledge item;

[0128] Each knowledge item is updated to the customer service robot knowledge base based on the confidence level of each knowledge item.

[0129] See also Figure 4 , Figure 4 Detailed description of an embodiment of a computer-readable storage medium provided by an embodiment of the present invention. Figure 4 As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, the following steps are implemented:

[0130] Acquire multi-source data; multi-source data includes user session log data, missed question data, and manual customer service answer data;

[0131] Perform problem semantic analysis on the missed problem data, and cluster the missed problem data based on the analysis results to obtain a set of unresolved problem types;

[0132] Based on the set of unresolved problem types, manual customer service answer data, and the customer service robot knowledge base, a knowledge association graph is constructed. The knowledge association graph is then matched with the set of unresolved problem types to obtain candidate knowledge entries for the problem.

[0133] For each candidate knowledge item in the question, each knowledge item is evaluated based on user session log data, manual customer service answer data, and the customer service robot knowledge base to obtain the confidence level of the knowledge item;

[0134] Each knowledge item is updated to the customer service robot knowledge base based on the confidence level of each knowledge item.

[0135] On the other hand, the present invention further provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the customer service robot knowledge base management method provided by the above methods, which includes:

[0136] Acquire multi-source data; multi-source data includes user session log data, missed question data, and manual customer service answer data;

[0137] Perform problem semantic analysis on the missed problem data, and cluster the missed problem data based on the analysis results to obtain a set of unresolved problem types;

[0138] Based on the set of unresolved problem types, manual customer service answer data, and the customer service robot knowledge base, a knowledge association graph is constructed. The knowledge association graph is then matched with the set of unresolved problem types to obtain candidate knowledge entries for the problem.

[0139] For each candidate knowledge item in the question, each knowledge item is evaluated based on user session log data, manual customer service answer data, and the customer service robot knowledge base to obtain the confidence level of the knowledge item;

[0140] Each knowledge item is updated to the customer service robot knowledge base based on the confidence level of each knowledge item.

[0141] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. That is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0142] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or certain parts of the embodiment.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for managing a customer service robot knowledge base, characterized in that: include: Acquire multi-source data; the multi-source data includes user session log data, missed question data, and manual customer service answer data; Performing problem semantic analysis on the missed problem data, and clustering the missed problem data based on the analysis result to obtain a set of unresolved problem types; Based on the set of unresolved problem types, the manual customer service answer data, and the customer service robot knowledge base, a knowledge association graph is constructed, and candidate problem knowledge entries are obtained by matching the knowledge association graph with the set of unresolved problem types; For each of the candidate knowledge items of the question, evaluate each knowledge item based on the user session log data, the manual customer service answer data, and the customer service robot knowledge base to obtain a confidence level of the knowledge item; Each of the knowledge items is updated to the customer service robot knowledge base based on the confidence level of each of the knowledge items.

2. The method for managing a customer service robot knowledge base according to claim 1, characterized in that: The step of performing semantic analysis on the missed question data and clustering the missed question data based on the analysis result to obtain a set of unresolved question types includes: Performing word segmentation processing on the missed question data to obtain a plurality of vocabulary data, and performing feature extraction on the vocabulary data based on a preset semantic knowledge base to obtain vocabulary semantic features; Performing syntactic structure analysis on the lexical semantic features to obtain syntactic structure information, and performing semantic role labeling on the lexical semantic features based on the syntactic structure information to obtain semantic role information; Performing semantic reasoning on the lexical semantic features, the syntactic structure information, and the semantic role information based on a preset semantic reasoning rule library to determine a semantic category; The missed question data are clustered using the lexical semantic features and the semantic categories to obtain a set of unresolved question types.

3. The method for managing a customer service robot knowledge base according to claim 2, characterized in that: The clustering process of the missed question data using the lexical semantic features and the semantic categories to obtain a set of unresolved question types includes: Constructing a semantic hierarchy tree based on the association relationship between the semantic categories and the lexical semantic features; Locating the unresolved issues in the missed issue data with the nodes in the semantic hierarchy tree to obtain unresolved issue nodes; Performing a path search on the unresolved problem nodes in the semantic hierarchy tree to obtain an unresolved problem node path, and performing in-depth processing on any two unresolved problem nodes based on a preset hierarchical correlation function to obtain a hierarchical correlation degree; Each unresolved problem in the missed problem data is used as an initial cluster, and the initial clusters are merged based on the hierarchical association to obtain the unresolved problem type set.

4. The method for managing a customer service robot knowledge base according to claim 1, characterized in that: The step of constructing a knowledge association graph based on the set of unresolved problem types, the manual customer service answer data, and the customer service robot knowledge base includes: Performing attribute extraction processing on the unresolved problems in the unresolved problem type set to obtain a problem attribute set; Classifying the problem attribute set to obtain problem attribute categories, and constructing a hierarchical structure between the problem attributes based on the hierarchical associations between the problem attributes in the problem attribute categories; Based on the hierarchical structure between the question attribute categories and the question attributes, matching is performed with the comprehensive knowledge in the customer service robot knowledge base and the manual customer service answer data to obtain question attribute knowledge entries; Analyzing the relationship between the problem attributes in the problem attribute knowledge items and constructing a problem attribute relationship model; Based on the question attribute relationship model, the knowledge association graph is constructed with question attributes as nodes and the relationships between question attributes as edges.

5. The method for managing a customer service robot knowledge base according to claim 4, characterized in that: The matching of the knowledge association graph with the set of unresolved problem types to obtain candidate problem knowledge entries includes: Decomposing the unresolved issues in the unresolved issue type set into semantic structures to obtain a plurality of semantic units; Mapping the semantic unit with the nodes in the knowledge association graph to obtain a semantic unit mapping node; Taking the semantic unit mapping node as a starting point, expanding the association path in the knowledge association graph to obtain an initial association path, and supplementing the initial association path based on the customer service robot knowledge base and an external knowledge source to obtain a target association path; For each unsolved problem in the unsolved problem type set, matching and screening the unsolved problem based on the target association path to obtain a knowledge entry; Each unresolved problem in the unresolved problem type set and its corresponding knowledge item are merged to obtain the candidate problem knowledge item.

6. The method for managing a customer service robot knowledge base according to claim 1, characterized in that: For each of the candidate knowledge items in the question, evaluating each knowledge item based on the user session log data, the manual customer service answer data, and the customer service robot knowledge base to obtain a confidence level corresponding to each knowledge item includes: Searching and matching paths related to the knowledge item in the knowledge association graph to obtain knowledge item path information; Search and match the information related to the knowledge item in the manual customer service answer data to obtain the knowledge item context information; Performing knowledge logic analysis on the knowledge item path information, and matching the logic analysis result with the knowledge item context information based on a matching function to obtain a first degree of fit; performing sentiment analysis on user conversation data related to the knowledge item in the user conversation log data, and performing a primary optimization on the first fit degree based on the sentiment analysis to obtain a second fit degree; An effect analysis is performed on the historical update feedback data related to the knowledge item in the customer service robot knowledge base, and the second fit is secondary optimized based on the effect analysis result to obtain the target fit, and the confidence level corresponding to each knowledge item is determined based on the target fit.

7. The method for managing a customer service robot knowledge base according to claim 6, characterized in that: The confidence levels include a first confidence level, a second confidence level, and a third confidence level. Based on the confidence level corresponding to each knowledge item, updating the knowledge item to the customer service robot knowledge base includes: If the confidence level corresponding to each of the knowledge items is the first confidence level, directly updating the knowledge item to the customer service robot knowledge base; If the confidence level corresponding to each of the knowledge items is the second confidence level, the knowledge item is updated to the customer service robot knowledge base after manual review and confirmation; If the confidence level corresponding to each of the knowledge items is the third confidence level, the characteristic information of the knowledge items is extracted, data mining is performed in multiple data sources based on the extracted characteristic information, and the mined data is evaluated for relevance. Based on the relevance evaluation results and the preset standard threshold, the data that meets the requirements is screened out, and then the screened multi-source data is fused. The knowledge items are updated and improved based on the fused data, and a secondary confidence evaluation is performed on the updated and improved knowledge items.

8. A management device for a customer service robot knowledge base, characterized in that: A method for managing a customer service robot knowledge base according to any one of claims 1 to 7; the management device for the customer service robot knowledge base comprises: A multi-source data acquisition module is used to obtain multi-source data; the multi-source data includes user session log data, missed question data, and manual customer service answer data; A semantic clustering analysis module, configured to perform semantic analysis on the missed question data, and cluster the missed question data based on the analysis results to obtain a set of unresolved question types; A knowledge association graph construction module is used to construct a knowledge association graph based on the set of unresolved problem types, the manual customer service answer data, and the customer service robot knowledge base, and to match the knowledge association graph with the set of unresolved problem types to obtain candidate problem knowledge entries; a confidence review module, configured to evaluate each knowledge item in the candidate knowledge items of the question based on the user session log data, the manual customer service answer data, and the customer service robot knowledge base to obtain a confidence level of the knowledge item; An update management module is used to update each knowledge item to the customer service robot knowledge base based on the confidence level of each knowledge item.

9. An electronic device comprising: Memory for storing computer software programs; A processor, configured to read and execute the computer software program, wherein when the processor executes the computer software program, it implements the method for managing the customer service robot knowledge base as claimed in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing a computer software program, wherein: When the computer software program is executed by a processor, the method for managing the customer service robot knowledge base as claimed in any one of claims 1 to 7 is implemented.

Images