Artificial intelligence robot language deep understanding method
By constructing feature models and integrating them into chatbots, the problem that chatbots in the prior art is difficult to distinguish business language from chat and identify business fields, and improve communication efficiency and accuracy.
Patent Information
- Application Number
- CN202411761135.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-03
- Publication Date
- 2025-03-11
AI Technical Summary
Existing chatbots cannot efficiently and accurately distinguish user input into business language or chat, and it is difficult to identify specific business areas for user consultation, resulting in inefficient communication during the company's office process.
Using the intent and entity understanding engine, data format conversion module, natural language processing module, machine learning module and deep learning module, a feature model is built to distinguish business languages from chat languages and identify specific business areas for user consultation.
It improves the accuracy of chatbots in distinguishing user input and identifying business fields, enhances the efficiency of communication during the business office process, and reduces the workload of repeated answers to questions.
Smart Images
Figure CN119597901A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of artificial intelligence natural language processing technology, and in particular to a method for deep language understanding of an artificial intelligence robot. Background Art
[0002] At present, the interactive mode of communication and dialogue in the process of office work in enterprises is mostly the traditional interactive mode of keyboard / mouse interface. It is impossible to optimize the interactive mode of communication and dialogue in the process of office work in the form of dialogue robots, and it is impossible to improve the previous interactive mode of enterprises through keyboard / mouse interface into the interactive mode of natural language, so as to help enterprises quickly build exclusive dialogue robots, and it is impossible for enterprises in all walks of life to realize the human-computer interaction capabilities of information management such as question and answer, tasks, tables, documents, reading comprehension, and knowledge graphs through natural language interaction. As a result, the workload of internal employees and external customer service for repeatedly answering questions and inquiries is greatly increased, and the overall efficiency of communication in the process of office work is reduced.
[0003] In order to solve the above technical problems, chatbots are now introduced in enterprise environments. These robots not only need to handle inquiries from different departments.
[0004] In addition, when it comes to quickly identifying the specific business areas of user consultation, current chatbots can only return answers to users based on their questions and matching their intentions. To determine whether the returned answers are the information the user needs, the user needs to ask again. Therefore, existing technologies usually cannot meet the dual needs of high efficiency and high accuracy. Summary of the invention
[0005] The present invention is intended to provide an artificial intelligence robot language deep understanding method, aiming to improve the accuracy of the chat robot in distinguishing user input as business language or small talk and identifying the specific business field of user consultation.
[0006] To achieve the above object, the present invention adopts the following technical solution: a method for deep language understanding of an artificial intelligence robot, comprising the following steps: Step 1: Collect and process user information: Use the intent and entity understanding engine to collect and identify user input information, and process it to obtain business language text information and chat language text information; Step 2: Text information format conversion: Use a data format conversion module to convert the business language text information and the chat language text information into data formats, and match them with the AI robot learning format; Step 3: Feature model construction and training: Combine the natural language processing module, machine learning module and deep learning module to build a basic model. In the basic model, extract features from business language text information and chat language text information to build a feature model. Use the random forest algorithm to train the feature model so that the feature model can distinguish between business language and chat language and identify the specific business areas of user consultation.
[0007] Furthermore, in step one, the intention and entity understanding engine includes the following steps: using intention matching to identify user intentions, identifying corresponding text information, performing Chinese word segmentation on the text information through a document preprocessing module, and establishing a text classification framework.
[0008] Furthermore, in step one, before the user inputs information, corpus is collected from different departments within the enterprise to construct a corpus, and the corpus is segmented and preprocessed in Chinese through a document preprocessing module to form a business segmentation corpus and a chat segmentation corpus.
[0009] In the present invention, in step three, before feature extraction and after Chinese word segmentation, the word vector processing module is used to generate word vectors for Chinese classification, and the word vector is the key link in feature model training, which ensures that text features can be effectively extracted.
[0010] Further, in step three, the feature model is trained and optimized through a machine learning module. The machine learning module extracts various features from the data sets of the language text in the business language text information and the chat language text information, sorts the various features according to importance, and removes a part of the features from the sorted features each time using the sequential backward selection method to obtain a new feature set after each round of screening, trains a random forest on the new feature set and calculates the corresponding accuracy; determines whether to continue screening based on the error increment caused by each round of screening relative to the error before screening, exits the iteration once it exceeds a specified threshold, and uses the feature set obtained from the previous round of screening as the result, repeats the operation, and finally uses the feature set with the highest test accuracy as the output result of feature selection.
[0011] Furthermore, in the machine learning module, a part of the features are removed from each sorted feature each time using the sequential backward selection method to obtain a new feature set after each round of screening, and simple cross-validation is used for screening.
[0012] Compared with the prior art, the present invention also has the following technical effects: By training corpora of different business fields, the model of the present invention can accurately identify the specific business field of user consultation, and integrate the trained model and rule engine into the chatbot, so that it can accurately identify the user's query intention and related fields in real-time conversations. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 The present invention is a flowchart of an artificial intelligence robot language deep understanding method.
Claims
1. A method for deep language understanding of an artificial intelligence robot, characterized in that: The following steps are involved: Step 1: Collect and process user information: Use the intent and entity understanding engine to collect and identify user input information, and process it to obtain business language text information and chat language text information; Step 2: Text information format conversion: Use a data format conversion module to convert the business language text information and the chat language text information into data formats, and match them with the AI robot learning format; Step 3: Feature model construction and training: Combine the natural language processing module, machine learning module and deep learning module to build a basic model. In the basic model, extract features from business language text information and chat language text information to build a feature model. Use the random forest algorithm to train the feature model so that the feature model can distinguish between business language and chat language and identify the specific business areas of user consultation.
2. The method for deep language understanding of an artificial intelligence robot according to claim 1, characterized in that: In step one, the intent and entity understanding engine includes the following steps: using intent matching to identify user intent, identifying corresponding text information, performing Chinese word segmentation on the text information through a document preprocessing module, and establishing a text classification framework; performing text extraction on the text information in the text classification framework, obtaining entity information in the text information, and constructing business language text information and chat language text information.
3. The method for deep language understanding of an artificial intelligence robot according to claim 1, characterized in that: In step 1, before the user inputs information, corpus is collected from different departments within the enterprise to construct a corpus, and the corpus is segmented and preprocessed by a document preprocessing module to form a business segmentation corpus and a chat segmentation corpus.
4. The method for deep language understanding of an artificial intelligence robot according to claim 1, characterized in that: In step three, the feature model is trained and optimized through the machine learning module. The machine learning module extracts various features from the data sets of language text in the business language text information and the chat language text information, sorts each feature according to importance, and removes a part of the features from the sorted features each time using the sequential backward selection method to obtain a new feature set after each round of screening, trains the random forest on the new feature set and calculates the corresponding accuracy; determines whether to continue screening based on the error increment caused by each round of screening relative to the error before screening, exits the iteration once it exceeds the specified threshold, and uses the feature set obtained in the previous round of screening as the result, repeats the operation, and finally uses the feature set with the highest test accuracy as the output result of feature selection.
5. The method for deep language understanding of an artificial intelligence robot according to claim 1, characterized in that: In the machine learning module, the sequential backward selection method is used each time to remove a part of the features from the sorted features, obtain a new feature set after each round of screening, and use simple cross-validation for screening.