Conversation method of intelligent customer service robot and knowledge base management system of intelligent customer service robot

Abstract

The invention relates to the technical field of customer service robots, and provides an intelligent customer service robot dialogue method and a knowledge base management system thereof.The intelligent customer service robot dialogue method comprises the steps that user intention recognition and information collection are achieved through a multi-round dialogue guiding module, and the module extracts keywords and analyzes sentence pattern structures through a natural language processing technology; if the intention is fuzzy, a question chasing mechanism is triggered to narrow the answering range, a standard answer is retrieved from a structured knowledge base through a knowledge base matching module on the basis of an intention recognition result, and the knowledge base comprises an FAQ base, a scene base and a case base and supports multi-mode output of texts, pictures, videos and the like. The intention of a user is accurately recognized through natural language processing and a questioning mechanism, and information collection is accelerated through an interactive menu and a form. A multi-modal knowledge base is combined with BERT semantic retrieval to improve the answer matching degree, and language style adaptation scenes are dynamically switched. The fuzzy matching algorithm judges the problem solving condition, and the process is automatically switched or restarted when the problem is not solved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of customer service robots, in particular to a dialogue method of an intelligent customer service robot and a knowledge base management system thereof. BACKGROUND

[0002] With the popularization of artificial intelligence technology in the customer service field, intelligent customer service robots have become the core tool for enterprises to reduce costs and increase efficiency. The existing scheme mainly realizes basic question and answer through keyword matching, but when facing multi-modal interaction, complex intent understanding and dynamic knowledge management requirements, the traditional system exposes significant limitations.

[0003] In the traditional customer service robot, the keyword matching cannot understand semantic association, multiple follow-up questions are needed to clarify the demand, the user interaction cost is increased, there is a lack of emotional analysis mechanism, the emotions such as anger and anxiety of the user cannot be recognized, the response strategy is single, the user satisfaction is low, the knowledge acquisition depends on manual input, the online period of new business knowledge is as long as 7 days, and the unstructured data cannot be automatically processed, the knowledge coverage is insufficient, the updating mechanism is lagging, the timely knowledge cannot be synchronized in real time, the outdated information leads to misguidance, and the enterprise compliance risk is significantly increased. SUMMARY

[0004] In view of the deficiencies of the prior art, the application provides a dialogue method of an intelligent customer service robot and a knowledge base management system thereof, which solves the problems that keyword matching cannot understand semantic association, multiple follow-up questions are needed to clarify the demand, and there is a lack of emotional analysis mechanism.

[0005] To achieve the above purpose, the application is implemented by the following technical scheme: a dialogue method of an intelligent customer service robot, comprising the following steps: Step 1: realizing user intent recognition and information collection through a multi-round dialogue guiding module, the module extracts keywords and analyzes sentence structure through natural language processing technology, and if the intent is ambiguous, a follow-up mechanism is triggered to narrow the answer range; Step 2: based on the intent recognition result, a knowledge base matching module is used to retrieve standard answers from a structured knowledge base, the knowledge base includes FAQ library, scene library and case library, and supports multi-modal output such as text, picture and video; Step 3: the result is confirmed and the satisfaction is investigated through a dialogue closed loop module, and the problem is automatically transferred to manual customer service or re-entered into the intent recognition process; Step 4: the emotional analysis unit is used to identify the user emotion in real time, and the response tone is adjusted in combination with a preset strategy, and the strategy includes an emotional soothing priority algorithm.

[0006] Preferably, the step 1 is a multi-round dialogue guiding module, which comprises: Opening unit: Provide an interactive greeting containing a functional menu, with response time meeting a preset threshold; Information extraction unit: Collect key information through structured forms or interactive buttons, supporting automatic filling of historical records.

[0007] Preferably, the second step is a knowledge base matching module, which includes: Semantic retrieval unit: Implement keyword and semantic mixed retrieval based on the BERT model, with retrieval accuracy meeting a preset threshold; Dynamic response unit: Automatically switch language styles according to business scenarios.

[0008] Preferably, the third step is a dialogue closed-loop module, which includes: Result confirmation unit: Use fuzzy matching algorithms to determine whether the user has solved the problem, with a matching threshold meeting a preset standard; Feedback optimization unit: Automatically update the knowledge base based on user evaluation data, with an update cycle meeting the business timeliness requirements.

[0009] A knowledge base management system for an intelligent customer service robot, comprising: Knowledge acquisition module: Collect multi-source data through data crawling sub-module, document parsing sub-module, and real-time data acquisition sub-module; Knowledge arrangement module: Includes data cleaning unit, knowledge classification unit, and knowledge graph generation unit; Knowledge update module: Predict update requirements through time series analysis sub-module, and implement incremental update of the knowledge base by combining a self-adaptive model update sub-module, with an update delay meeting real-time requirements; Knowledge security module: Use the Advanced Encryption Standard to field-level encrypt sensitive data, and implement hierarchical management of permissions through role-based access control.

[0010] Preferably, the knowledge acquisition module includes: Real-time data processing unit: Preliminarily analyze real-time data collected by sensors based on stream processing technology; Data verification unit: Eliminate invalid data through machine learning anomaly detection algorithms.

[0011] Preferably, the knowledge classification unit of the knowledge arrangement module uses a hybrid model of convolutional neural networks and rule engines to hierarchically classify knowledge and supports manual adjustment of classification labels.

[0012] Preferably, the knowledge update module includes: Version management unit: Record the modification history of knowledge entries, supporting version rollback; Time-sensitive reminder unit: Automatically trigger an update reminder for knowledge entries with an effective period less than 7 days.

[0013] Preferably, the knowledge security module comprises: Access audit unit: real-time record user operation log; Emergency response unit: automatically trigger the fuse mechanism when detecting abnormal access.

[0014] Preferably, it also includes a knowledge recommendation module, which comprises: Collaborative filtering submodule: generate recommendation list based on user historical behavior data; Deep learning recommendation submodule: generate personalized recommendation through Transformer model combined with context.

[0015] The present application provides a dialogue method of intelligent customer service robot and its knowledge base management system. With the following beneficial effects: 1. The present application accurately identifies user intent through natural language processing and follow-up mechanism, and interactive menu and form accelerate information collection. Multi-modal knowledge base combined with BERT semantic retrieval improves answer matching degree, and dynamically switches language style to adapt to the scene. Fuzzy matching algorithm determines the problem solving situation, and automatically transfers or restarts the process for unsolved problems. User evaluation drives dynamic optimization of knowledge base, shortening the service cycle.

[0016] 2. The present application analyzes user emotions in real time through the model, and adjusts the response tone according to the priority triggered by the pacification technique. The knowledge recommendation module generates personalized content based on historical behavior and context, such as product matching recommendation. In the scene of big promotion, real-time information is automatically synchronized and the language is switched, and the timeliness and emotional resonance of scenario-based service are strengthened.

[0017] 3. The present application integrates real-time business information through multi-source data collection, and ensures data validity through stream processing and anomaly detection. Convolutional neural network and rule engine hybrid classification knowledge, knowledge graph improves correlation retrieval ability. Dynamic update mechanism combined with version management and time limit reminder maintains the timeliness of content, and field encryption, permission control and audit mechanism guarantee data security. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The flowchart of the present application; Figure 2 The system diagram of the present application. DETAILED DESCRIPTION

[0019] 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 part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] Embodiments: Please refer to the accompanying Figure 1 - the accompanying Figure 2 The embodiments of the present application provide a dialogue method of an intelligent customer service robot, comprising the following steps: Step one: realize user intent recognition and information collection through a multi-round dialogue guide module, the module extracts keywords and analyzes sentence structure through natural language processing technology, if the intent is ambiguous, trigger the follow-up mechanism to narrow down the answer range, step one is a multi-round dialogue guide module, which comprises: Opening speech unit: provide an interactive greeting containing a function menu, the menu response time meets the preset threshold, generate a visual menu based on the front-end interaction framework, integrate speech synthesis technology to broadcast the greeting simultaneously, the menu response time is short to meet the user's immediate interaction needs; Information extraction unit: collect key information through structured forms or interactive buttons, support automatic filling of historical records, use JSON format structured forms, support OCR automatic recognition of user uploaded picture text, combine historical dialogue records to automatically fill in known information; Step two: based on the intent recognition result, retrieve standard answers from the structured knowledge base through the knowledge base matching module, the knowledge base includes FAQ library, scenario library and case library, supports multi-modal output such as text, picture and video, step two is a knowledge base matching module, which comprises: Semantic retrieval unit: realize keyword and semantic hybrid retrieval based on BERT model, the retrieval accuracy meets the preset threshold, construct a semantic vector space based on the BERT pre-training model, calculate the cosine similarity between user questions and knowledge base items combined with the TF-IDF algorithm, support fuzzy query; Dynamic response unit: automatically switch language style according to business scenarios, preset scenario labels, automatically switch language style through rule engine; Step three: confirm the result and investigate the satisfaction through the dialogue closed loop module, automatically transfer to manual customer service or re-enter the intent recognition process if the problem is not solved, step three is a dialogue closed loop module, which comprises: Result confirmation unit: determine whether the user has solved the problem using fuzzy matching algorithm, the matching threshold meets the preset standard, analyze the user's reply text using fuzzy matching algorithm, and determine that the problem is solved when the matching threshold is greater than or equal to 0.8; combined with voice emotion recognition; Feedback optimization unit: automatically update the knowledge base based on user evaluation data, the update cycle meets the business timeliness requirements, determine the polarity of user evaluation through NLP sentiment analysis model, and automatically mark low-score items as to be optimized; Step four: Real-time recognition of user emotions using the sentiment analysis unit, combined with preset response strategies, including an emotion pacification priority algorithm, using an LSTM-Attention model to analyze emotional features in speech or text, and a preset priority algorithm, such as prioritizing angry users to trigger pacification phrases like "I'm very sorry for the inconvenience, we will handle it immediately."

[0021] A knowledge base management system for an intelligent customer service robot, comprising: A knowledge acquisition module that collects multi-source data through data crawling, document parsing, and real-time data acquisition sub-modules, including: A real-time data processing unit that performs preliminary analysis on real-time data collected by sensors based on stream processing technology, processes sensor data based on the Flink stream processing framework, and uses a sliding window algorithm to real-time statistics of abnormal indicators; A data verification unit that uses machine learning anomaly detection algorithms to eliminate invalid data and uses the IsolationForest algorithm to detect data outliers and eliminate invalid data.

[0022] A knowledge refinement module that includes a data cleaning unit, a knowledge classification unit, and a knowledge graph generation unit, the knowledge classification unit of the knowledge refinement module uses a hybrid model of convolutional neural networks and rule engines for hierarchical classification of knowledge, and supports manual adjustment of classification labels, the data cleaning unit can remove random codes and special symbols in text, and uses a data deduplication algorithm to identify similar entries, the knowledge classification unit extracts text features and uses a rule engine for hierarchical classification, supports manual adjustment of labels, and the knowledge graph generation unit uses a Neo4j graph database to construct a knowledge graph based on entity relationship extraction.

[0023] A knowledge update module that predicts update needs through a time series analysis sub-module and implements incremental updates to the knowledge base using an adaptive model update sub-module, with update delays meeting real-time requirements, including: A version management unit that records the modification history of knowledge items and supports version rollback, using Git version control technology to record the modification history of knowledge items and supporting one-click rollback to the previous version; A time-sensitive reminder unit that automatically triggers update reminders for knowledge items with an effective period of less than 7 days, sets an effective period for knowledge items, and automatically sends emails or SMS messages to administrators 7 days before expiration.

[0024] A knowledge security module that uses the Advanced Encryption Standard to encrypt sensitive data at the field level and implements hierarchical access control through role-based access control, including: An access audit unit that records user operation logs in real-time; Emergency response unit: automatically trigger the fuse mechanism when detecting abnormal access, such as when discovering attack behaviors such as brute force cracking, and send an alarm to the security team.

[0025] Also includes a knowledge recommendation module, which includes: Collaborative filtering submodule: generate a recommendation list based on user historical behavior data, analyze user historical consultation records, generate a similar user group through matrix decomposition algorithm, and recommend high-frequency access knowledge of the group; Deep learning recommendation submodule: generate personalized recommendations through the Transformer model combined with the context, build a user interest vector based on the Transformer model, and generate a real-time recommendation list combined with the current dialogue context.

[0026] Embodiment 1: After the user scans the product two-dimensional code, the robot collects preference information such as size and color through multiple rounds of dialogue guidance, and pushes matching products in combination with the knowledge recommendation module to improve conversion rate. During the big promotion period, the knowledge base automatically synchronizes real-time inventory, logistics anomalies and other information, and switches to promotion-specific language style through the dynamic response unit.

[0027] Embodiment 2: When handling "credit card theft" consultations, the sentiment analysis unit identifies user anxiety, prioritizes human customer service, and associates knowledge such as "emergency loss process" and "transaction dispute handling policy" through the knowledge graph to shorten processing time. The knowledge update module regularly scrapes central government policy documents and automatically updates the compliance knowledge base through the document parsing submodule to ensure that the answers meet the latest regulatory requirements.

[0028] Embodiment 3: The real-time data processing unit analyzes abnormal logs reported by the smart speaker, automatically generates fault cases and pushes them to the after-sales knowledge base, and maintenance personnel can quickly obtain solutions through semantic retrieval. When the user consults "device networking failure", the system associates "router model X compatibility problem" and "firmware upgrade tutorial" through the knowledge graph to provide multi-modal answers in the form of text and video.

[0029] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A dialogue method for an intelligent customer service robot, characterized in that, Includes the following steps: Step 1: User intent recognition and information collection are achieved through a multi-turn dialogue guidance module. This module extracts keywords and analyzes sentence structure using natural language processing technology. If the intent is ambiguous, a follow-up questioning mechanism is triggered to narrow down the range of answers. Step 2: Based on the intent recognition results, the standard answer is retrieved from the structured knowledge base through the knowledge base matching module. The knowledge base includes a FAQ library, a scenario library, and a case library, and supports multimodal output such as text, images, and videos. Step 3: Confirm the results and conduct a satisfaction survey through the dialogue loop module. Unresolved issues will be automatically transferred to human customer service or the process will re-enter the intent recognition process. Step 4: Use the sentiment analysis unit to identify user emotions in real time, and adjust the response tone in combination with preset strategies, including an emotion soothing priority algorithm.

2. The dialogue method for an intelligent customer service robot according to claim 1, characterized in that, Step one is a multi-turn dialogue guidance module, which includes: Opening Remarks Unit: Provides an interactive greeting with a function menu, and the menu response time meets a preset threshold; Information extraction unit: Collects key information through structured forms or interactive buttons, and supports automatic filling of history records.

3. The dialogue method for an intelligent customer service robot according to claim 1, characterized in that, Step two is a knowledge base matching module, which includes: Semantic retrieval unit: Based on the BERT model, it realizes hybrid keyword and semantic retrieval, and the retrieval accuracy meets the preset threshold; Dynamic response unit: Automatically switches language style according to business scenarios.

4. The dialogue method for an intelligent customer service robot according to claim 1, characterized in that, Step three is the dialogue loop closure module, which includes: Result confirmation unit: Uses fuzzy matching algorithm to determine whether the user has solved the problem, and the matching threshold meets the preset standard; Feedback optimization unit: Automatically updates the knowledge base based on user evaluation data, with an update cycle that meets business timeliness requirements.

5. A knowledge base management system for an intelligent customer service robot, using the dialogue method of an intelligent customer service robot as described in any one of claims 1-4, characterized in that, include: Knowledge acquisition module: Collects multi-source data through data crawling submodule, document parsing submodule and real-time data acquisition submodule; The knowledge organization module includes a data cleaning unit, a knowledge classification unit, and a knowledge graph generation unit. Knowledge update module: It predicts update needs through time series analysis submodule and realizes incremental update of knowledge base by combining adaptive model update submodule, and the update delay meets the real-time requirements; Knowledge security module: It uses advanced encryption standards to encrypt sensitive data at the field level and implements hierarchical access control through role-based access control.

6. The knowledge base management system for an intelligent customer service robot according to claim 5, characterized in that, The knowledge acquisition module includes: Real-time data processing unit: performs preliminary analysis on real-time data acquired by sensors based on stream processing technology; Data validation unit: Removes invalid data using machine learning anomaly detection algorithms.

7. The knowledge base management system for an intelligent customer service robot according to claim 5, characterized in that, The knowledge classification unit of the knowledge organization module adopts a hybrid model of convolutional neural network and rule engine to classify knowledge hierarchically and supports manual adjustment of classification labels.

8. The knowledge base management system for an intelligent customer service robot according to claim 5, characterized in that, The knowledge update module includes: Version management unit: Records the modification history of knowledge entries and supports version rollback; Time-sensitive reminder unit: Automatically triggers update reminders for knowledge items with a validity period of less than 7 days.

9. The knowledge base management system for an intelligent customer service robot according to claim 5, characterized in that, The knowledge security module includes: Access audit unit: Records user operation logs in real time; Emergency Response Unit: Automatically triggers the circuit breaker mechanism when abnormal access is detected.

10. A knowledge base management system for an intelligent customer service robot according to claim 5, characterized in that, It also includes a knowledge recommendation module, which includes: Collaborative filtering submodule: Generates a recommendation list based on users' historical behavior data; Deep learning recommendation submodule: Generates personalized recommendations by combining the Transformer model with context.

Images