Intelligent interaction system based on Chinese natural language processing

Abstract

The invention provides an intelligent interaction system based on Chinese natural language processing, and relates to the technical field of artificial intelligence, and the system comprises the steps: employing a natural language processing model obtained through pre-training to carry out word segmentation on the dialogue content of a user to obtain a word segmentation result, and carrying out part-of-speech tagging and named entity recognition on the word segmentation result to obtain the semantics of the user; judging the context state of the current dialogue based on semantics, querying according to the context state to obtain related knowledge information and determine a response strategy, and keeping the context of the current dialogue; and generating reply text according to the context state and the related knowledge information and based on the response strategy. The method has the beneficial effects that the Chinese language is deeply optimized, so that the expression is more accurate in ambiguity processing, context association and oral expression understanding; deep optimization is carried out aiming at language habits and cultural backgrounds of Chinese users, the method has unique advantages in understanding and responding to Chinese expression, and more localized user experience is provided.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intelligent interaction system based on Chinese natural language processing. Background Art

[0002] With the rapid development of artificial intelligence technology, chatbots have become an important tool widely used in all walks of life. They provide users with convenient services in customer service, education, medical care and other fields, greatly improving work efficiency and user experience. However, most of the current chatbot systems are developed based on English or other Western language environments, which leads to certain limitations in their performance in Chinese environments.

[0003] As a unique language, Chinese has its own complexity and richness. In addition, Chinese has a huge vocabulary, many of which have multiple meanings and may have different interpretations in different contexts. For example, when a user mentions "moon", it may involve poetry, science, or even personal emotions in different contexts. Chatbots often fail to meet user needs if they rely solely on literal meanings. The close integration of cultural background is also an important factor affecting the performance of chatbots in Chinese environments. China has a long history and rich culture, and many social customs, etiquette, and values ​​are deeply rooted in people's daily communication. Therefore, when designing chatbots, these cultural factors must be fully considered in order to better communicate with users. For example, in Chinese social situations, politeness and humility are very important qualities, and chatbots should reflect this when responding to users instead of simply giving direct answers.

[0004] Currently, many chatbots still have significant deficiencies in managing multi-turn conversations. Multi-turn conversations refer to continuous interactions between users and robots, in which the robot needs to be able to remember the content of previous conversations and make reasonable responses based on this information. However, many existing systems are weak in this regard and are often only able to handle simple questions and answers, but are unable to conduct in-depth conversations.

[0005] In addition to traditional text interaction, modern chatbots also need to support multimodal interaction. This means that in addition to text, the robot should also be able to understand other forms of information such as voice and images. However, most chatbots currently still rely mainly on text input and lack effective processing of other forms of input. This is particularly prominent in the Chinese environment because many users are accustomed to communicating through voice input, especially on mobile devices.

[0006] As technology develops and user needs continue to evolve, users' expectations of chatbots are also gradually increasing. Users hope that chatbots can not only answer questions quickly and accurately, but also have more natural conversations and understand their emotions and intentions. Therefore, existing chatbot systems urgently need to be upgraded to meet these growing needs. Summary of the invention

[0007] In view of the problems existing in the prior art, the present invention provides an intelligent interactive system based on Chinese natural language processing, comprising:

[0008] A natural language processing module, used to segment the user's conversation content using a pre-trained natural language processing model to obtain a segmentation result, and to perform part-of-speech tagging and named entity recognition on the segmentation result to obtain the user's semantics;

[0009] Knowledge base module, used to store pre-built knowledge information;

[0010] A dialogue management module, connected to the natural language processing module and the knowledge base module respectively, for determining the context state of the current dialogue based on the semantics, obtaining the relevant knowledge information and determining a response strategy according to the context state, while maintaining the context of the current dialogue;

[0011] A response generation module is connected to the dialog management module and is used to generate a reply text according to the context state and the related knowledge information and based on the response strategy.

[0012] Preferably, the natural language processing module includes a word segmentation submodule, which is used to input the conversation content into a pre-trained word segmentation model for word segmentation processing, so as to identify keywords and sentence structures contained in the conversation content as the word segmentation results;

[0013] The word segmentation model includes a bidirectional long short-term memory network, the input of the bidirectional long short-term memory network is the input of the word segmentation model, the output of the bidirectional long short-term memory network is connected to a conditional random field layer, and the output of the conditional random field layer is the output of the word segmentation model.

[0014] Preferably, the semantics includes the user's conversation intention, and the natural language processing module includes an intention recognition submodule, including:

[0015] A first recognition unit is used to process the word segmentation result according to a pre-trained part-of-speech tagging and entity recognition model to obtain the part-of-speech of the keyword and the named entity contained therein;

[0016] The second recognition unit is connected to the first recognition unit and is used to perform intent recognition according to the part of speech of the keyword and the named entities contained therein to obtain the conversation intention of the user.

[0017] Preferably, the part-of-speech tagging and entity recognition model is obtained by self-supervised learning and transfer learning training.

[0018] Preferably, it also includes a data monitoring module connected to the knowledge base module, which is used to monitor the real-time dynamics of each data source that provides the knowledge information in real time, and update the relevant knowledge information in real time when the real-time dynamics indicate that there is a data change event in the data source.

[0019] Preferably, the data monitoring module includes a data selection submodule, which is used to use a data fusion algorithm based on a graph neural network to eliminate duplicate change data and select the optimal change data to update the related knowledge information when the data change event is monitored and the change data corresponding to the data change event comes from multiple data sources at the same time.

[0020] Preferably, it also includes a user interface module, connected to the natural language processing module, for allowing the user to enter the conversation content.

[0021] Preferably, the natural language processing module includes a conversion submodule connected to the word segmentation submodule, and configured to convert the speech into text using a pre-trained speech recognition model when the conversation content of the user is speech, and send the text to the word segmentation submodule;

[0022] The speech recognition model includes one or more layers of convolutional neural networks, the input of the convolutional neural network is the input of the speech recognition model, the output of the convolutional neural network is connected to a fully connected layer, and the output of the fully connected layer is the output of the speech recognition model.

[0023] Preferably, it also includes a learning and feedback module, which is connected to the user interface module and is used to record the user's dialogue feedback information after each dialogue ends, and optimize the natural language processing model and the dialogue strategy based on the dialogue feedback information and using a machine learning algorithm.

[0024] Preferably, the response generation module includes a text conversion submodule, which is used to convert the reply text into a voice response and return it to the user when the conversation content entered by the user is in voice format.

[0025] The above technical solution has the following advantages or beneficial effects:

[0026] 1) Chinese semantic understanding: Through in-depth optimization of the Chinese language, it is more accurate in polysemy processing, contextual association, and oral expression understanding;

[0027] 2) Self-learning mechanism: Continuously monitor the update of data sources. Through dynamic knowledge base, learning and feedback, the system can continuously optimize according to user feedback, improve the accuracy of dialogue and the overall performance of the system;

[0028] 3) Localization advantage: It has been deeply optimized based on the language habits and cultural background of Chinese users, has unique advantages in understanding and responding to Chinese expressions, and provides a more localized user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 In a preferred embodiment of the present invention, an intelligent interactive system based on Chinese natural language processing is provided. DETAILED DESCRIPTION

[0030] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The present invention is not limited to this embodiment, and other embodiments may also fall within the scope of the present invention as long as they conform to the gist of the present invention.

[0031] In a preferred embodiment of the present invention, based on the above problems existing in the prior art, an intelligent interactive system based on Chinese natural language processing is provided. Figure 1 As shown, including:

[0032] A natural language processing module 1 is used to segment the user's conversation content using a pre-trained natural language processing model to obtain a segmentation result, and to perform part-of-speech tagging and named entity recognition on the segmentation result to obtain the user's semantics;

[0033] Knowledge base module 2, used to store pre-built knowledge information;

[0034] The dialogue management module 3 is connected to the natural language processing module 1 and the knowledge base module 2 respectively, and is used to determine the context state of the current dialogue based on semantics, obtain relevant knowledge information and determine the response strategy according to the context state query, and at the same time maintain the context of the current dialogue;

[0035] The response generation module 4 is connected to the dialogue management module 3 and is used to generate a reply text according to the context state and related knowledge information and based on the response strategy.

[0036] Specifically, the intelligent interactive system of the present invention is applied to an intelligent customer service system of an enterprise as an example for explanation:

[0037] In this embodiment, the intelligent interactive system of the present invention also provides a user interface module 6, which is connected to the natural language processing module 1 and is used for users to enter conversation content. The user interface module 6 supports interactive interfaces of multiple devices and platforms, and users can have conversations through mobile devices, web pages or other application platforms. In this embodiment, it can be preferably deployed in the official website and mobile applications of the enterprise. Users can enter conversation content through the chat window provided by the user interface module 6, and the user interface module 6 is responsible for passing the conversation content entered by the user to the natural language processing module 1 for subsequent processing.

[0038] Among them, the user can input text or voice through the chat window. After receiving the input content, the natural language processing module 1 needs to pre-process the input content first. There are two cases at this time:

[0039] In one case, the input content is text, then the natural language processing module 1 includes a word segmentation submodule 11, which is used to input the conversation content into a pre-trained word segmentation model for word segmentation processing, so as to identify the keywords and sentence structures contained in the conversation content as the word segmentation results;

[0040] The word segmentation model includes a bidirectional long short-term memory network, the input of the bidirectional long short-term memory network is the input of the word segmentation model, the output of the bidirectional long short-term memory network is connected to the conditional random field layer, and the output of the conditional random field layer is the output of the word segmentation model.

[0041] Specifically, the word segmentation model uses an algorithm based on a bidirectional long short-term memory network (Bi-LSTM) combined with a conditional random field (CRF), which is optimized specifically for the lexical characteristics of Chinese. Compared with traditional word segmentation methods based on rules or simple statistics, the word segmentation model improves the accuracy of word segmentation through context and character proximity, especially for the recognition of polysemous words and new words, effectively dealing with the complex structure of Chinese.

[0042] After obtaining keywords and sentence structures through word segmentation, it is necessary to further analyze the text content, determine the user's intention, and understand the deep meaning of the user's input in context. The deep meaning understanding here includes its meaning in the current context.

[0043] Furthermore, the semantics includes the user's conversation intention, and the natural language processing module 1 includes an intention recognition submodule 12, including:

[0044] The first recognition unit 121 is used to process the word segmentation result according to the pre-trained part-of-speech tagging and entity recognition model to obtain the part-of-speech of the keyword and the named entity contained;

[0045] The second recognition unit 122 is connected to the first recognition unit 121 and is used to perform intent recognition according to the part of speech of the keyword and the included named entities to obtain the user's conversation intent.

[0046] Specifically, the word segmentation results are tagged with parts of speech and named entity recognition is performed to identify intent, where the intent includes querying order status or requesting help, etc. The results of the above part of speech tagging include nouns, verbs, adjectives, etc., and the above identified named entities include but are not limited to names of people, places, and institutions.

[0047] In a preferred embodiment of the present invention, self-supervised learning and transfer learning training are used to obtain a part-of-speech tagging and entity recognition model, which can quickly adapt to changes in different fields and new vocabulary based on the unique language structure and grammatical rules of Chinese, thereby improving recognition accuracy.

[0048] After identifying the user's intention, it is sent to the dialogue management module 3. The dialogue management module 3 determines the response strategy based on the context and user intention and the pre-built knowledge base module 2. In other words, the dialogue management module 3 can call the knowledge base module 2 or other external resources for query and reasoning to give a response strategy. In short, the dialogue management module 3 can judge the context state of the current dialogue based on the output of the natural language processing module 1 and determine the next response strategy. For example, the user asks "Has my order been shipped?" The natural language processing module 1 will recognize that the intention belongs to an order query request. The dialogue management module 3 obtains relevant knowledge information by querying the knowledge base module 2 and maintains the context of the current dialogue in order to handle further questions that may be raised by subsequent users.

[0049] In this embodiment, the knowledge information in the knowledge base module 2 includes but is not limited to answers to frequently asked questions, order information database, and user account information. It can be understood that if the intelligent interactive system of the present invention is not applied to the intelligent customer service system of an enterprise, the above knowledge information must also be adaptively adjusted based on the application scenario.

[0050] In a preferred embodiment of the present invention, it also includes a data monitoring module 5, which is connected to the knowledge base module 2 and is used to monitor the real-time dynamics of various data sources that provide knowledge information in real time, and update the relevant knowledge information in real time when the real-time dynamics indicate that there is a data change event in the data source.

[0051] Specifically, the above data change events include new events, modification events and deletion events of the data source. Once the data monitoring module 5 detects that the data source has changed due to new product releases or policy changes, it can automatically trigger the update process of the knowledge information. Compared with the manual or periodic batch update of the knowledge base in the prior art, the present invention can update immediately when the data source changes through an automated monitoring and triggering mechanism, realize the real-time dynamic update of the knowledge information, and ensure that the knowledge base is always kept up to date.

[0052] Furthermore, the data monitoring module 5 includes a data selection submodule 51, which is used to use a data fusion algorithm based on a graph neural network to eliminate duplicate change data and select the optimal change data to update related knowledge information when a data change event is monitored and the change data corresponding to the data change event comes from multiple data sources at the same time.

[0053] Specifically, compared with the traditional knowledge base system, which has limited processing capabilities for multiple data sources and is prone to redundancy and conflict, in this embodiment, by introducing a data fusion algorithm based on a graph neural network, it is possible to effectively eliminate duplicate information and select the optimal data, thereby improving the accuracy and completeness of the data.

[0054] After the dialogue management module 3 outputs the response strategy, the response generation module 4 generates a reply text according to the context state and related knowledge information and based on the response strategy, and then returns it to the user through the user interface module 6. The user can choose to continue the interaction or end the dialogue based on the reply text, thus completing a round of dialogue process.

[0055] In another case, the input content is voice, then the natural language processing module 1 includes a conversion submodule 13, connected to the word segmentation submodule 11, for converting the voice into text using a pre-trained voice recognition model and sending the text to the word segmentation submodule when the user's conversation content is voice;

[0056] The speech recognition model includes one or more layers of convolutional neural networks. The input of the convolutional neural network is the input of the speech recognition model. The output of the convolutional neural network is connected to the fully connected layer, and the output of the fully connected layer is the output of the speech recognition model.

[0057] Specifically, if the input content is speech, the speech is first converted into text through speech recognition, and then word segmentation and other operations are performed. The specific operation process is the same as above and will not be repeated here. Among them, the speech recognition model is based on deep neural network and convolutional neural network technology, combined with the characteristics of Chinese pronunciation, to perform efficient speech recognition. Compared with the existing recognition model, it can better recognize the unique tones and polyphonic characters of Chinese, and reduce common errors in Chinese speech recognition.

[0058] Furthermore, if the input content is voice, it is considered that the user prefers voice communication. To meet user needs, in a preferred embodiment of the present invention, the response generation module 4 includes a text conversion submodule 41, which is used to convert the reply text into a voice response and return it to the user when the conversation content input by the user is in voice format.

[0059] In a preferred embodiment of the present invention, it also includes a learning and feedback module 7, which is connected to the user interface module 6 and is used to record the user's dialogue feedback information after each dialogue ends, and optimize the natural language processing model and dialogue strategy based on the dialogue feedback information and using a machine learning algorithm.

[0060] Specifically, in this embodiment, after each conversation, the learning and feedback module 7 records the user's conversation feedback information, and then the system can use this conversation feedback information and its learning algorithm to continuously optimize the natural language processing model and conversation strategy, so that the system can perform more accurately and efficiently in future conversations.

[0061] As a preferred embodiment of the present invention, the operation flow of the system is described by taking a customer service scenario as an example:

[0062] Step 1: User Input

[0063] The user enters "I want to check the status of my order" through the chat window of the official website.

[0064] (Note: Users can input text or voice, and the system will automatically recognize it.)

[0065] Step 2: Natural Language Processing

[0066] The system's natural language processing module (NLP) receives the user's input and performs word segmentation, intent recognition and semantic understanding.

[0067] The word segmentation result may be: "I|want|to|query|the|status|of|my|order".

[0068] Intent recognition: Determine the user's intention as "query order status".

[0069] Semantic understanding: Combined with the context, it understands that the user wants to obtain the shipping status of a specific order.

[0070] Step 3: Dialogue Management

[0071] The dialog management module determines that the status of the current dialog is "order query" and calls the knowledge base module to retrieve the user's order information.

[0072] Step 4: Knowledge Base Query

[0073] The system queries the user's order information database, finds the order record related to the current user, and extracts its delivery status.

[0074] Step 5: Response Generation

[0075] After the knowledge base module returns the order status information, the response generation module generates the answer text: "Your order was shipped yesterday and is expected to arrive within three days."

[0076] Step 6: User feedback and system optimization

[0077] The user is satisfied with the answer and gives positive feedback. The learning and feedback module records the relevant data of this conversation and uses it for model optimization to improve the processing efficiency of similar conversations in the future.

[0078] It can be seen that each step depends on the output of the previous step, and the order cannot be adjusted arbitrarily. For example, user input must be processed by the natural language processing module before it can be decided by the dialogue management module, and only after the decision is made can the knowledge base module be queried. Therefore, the order of steps is strictly carried out according to the following process: user input → natural language processing → dialogue management → knowledge base query → response generation → user feedback and system optimization.

[0079] In summary, the intelligent interactive system of the present invention has significant advantages in the following aspects:

[0080] 1) Chinese semantic understanding: Through in-depth optimization of the Chinese language, it is more accurate in polysemy processing, contextual association, and oral expression understanding;

[0081] 2) Self-learning mechanism: Continuously monitor the update of data sources. Through dynamic knowledge base, learning and feedback, the system can continuously optimize according to user feedback, improve the accuracy of dialogue and the overall performance of the system;

[0082] 3) Localization advantage: It has been deeply optimized based on the language habits and cultural background of Chinese users, has unique advantages in understanding and responding to Chinese expressions, and provides a more localized user experience.

[0083] The above description is only a preferred embodiment of the present invention, and does not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of this specification and illustrations should be included in the protection scope of the present invention.

Claims

1. An intelligent interactive system based on Chinese natural language processing, characterized in that: include: A natural language processing module, used to segment the user's conversation content using a pre-trained natural language processing model to obtain a segmentation result, and to perform part-of-speech tagging and named entity recognition on the segmentation result to obtain the user's semantics; Knowledge base module, used to store pre-built knowledge information; A dialogue management module, connected to the natural language processing module and the knowledge base module respectively, for determining the context state of the current dialogue based on the semantics, obtaining the relevant knowledge information and determining a response strategy according to the context state, while maintaining the context of the current dialogue; A response generation module is connected to the dialog management module and is used to generate a reply text according to the context state and the related knowledge information and based on the response strategy.

2. The intelligent interactive system according to claim 1, characterized in that: The natural language processing module includes a word segmentation submodule, which is used to input the conversation content into a pre-trained word segmentation model for word segmentation processing, so as to identify keywords and sentence structures contained in the conversation content as the word segmentation results; The word segmentation model includes a bidirectional long short-term memory network, the input of the bidirectional long short-term memory network is the input of the word segmentation model, the output of the bidirectional long short-term memory network is connected to a conditional random field layer, and the output of the conditional random field layer is the output of the word segmentation model.

3. The intelligent interactive system according to claim 2, characterized in that: The semantics includes the user's conversation intention, and the natural language processing module includes an intention recognition submodule, including: A first recognition unit is used to process the word segmentation result according to a pre-trained part-of-speech tagging and entity recognition model to obtain the part-of-speech of the keyword and the named entity contained therein; The second recognition unit is connected to the first recognition unit and is used to perform intent recognition according to the part of speech of the keyword and the named entities contained therein to obtain the conversation intention of the user.

4. The intelligent interactive system according to claim 3, characterized in that: The part-of-speech tagging and entity recognition model is obtained by adopting self-supervised learning and transfer learning training.

5. The intelligent interactive system according to claim 1, characterized in that: It also includes a data monitoring module, which is connected to the knowledge base module and is used to monitor the real-time dynamics of each data source that provides the knowledge information in real time, and update the relevant knowledge information in real time when the real-time dynamics indicate that there is a data change event in the data source.

6. The intelligent interactive system according to claim 5, characterized in that: The data monitoring module includes a data selection submodule, which is used to use a data fusion algorithm based on a graph neural network to eliminate duplicate change data and select the optimal change data to update the related knowledge information when the data change event is monitored and the change data corresponding to the data change event comes from multiple data sources at the same time.

7. The intelligent interactive system according to claim 1, characterized in that: It also includes a user interface module, which is connected to the natural language processing module and is used for the user to input the conversation content.

8. The intelligent interactive system according to claim 2, characterized in that: The natural language processing module includes a conversion submodule connected to the word segmentation submodule, which is used to convert the speech into text using a pre-trained speech recognition model and send it to the word segmentation submodule when the conversation content of the user is speech; The speech recognition model includes one or more layers of convolutional neural networks, the input of the convolutional neural network is the input of the speech recognition model, the output of the convolutional neural network is connected to a fully connected layer, and the output of the fully connected layer is the output of the speech recognition model.

9. The intelligent interactive system according to claim 7, characterized in that: It also includes a learning and feedback module, which is connected to the user interface module and is used to record the user's dialogue feedback information after each dialogue, and optimize the natural language processing model and the dialogue strategy based on the dialogue feedback information and using a machine learning algorithm.

10. The intelligent interactive system according to claim 1, characterized in that: The response generation module includes a text conversion submodule, which is used to convert the reply text into a voice response and return it to the user when the conversation content entered by the user is in voice format.

Images