Virtual human intelligent answering system based on Internet

Abstract

The invention belongs to the technical field of intelligent answering, and particularly relates to an internet-based virtual human intelligent answering system, which comprises a virtual human terminal used for intelligent voice and image interaction; the virtual human terminal comprises a power supply module, an input module, a processing unit and an output module; the processing unit is internally provided with a CPU, and a voice recognition module, an answer generation module, a relation inference module, a keyword recognition module, a knowledge base storage module and a field expansion module which are in electric signal connection with the CPU. The voice recognition module further recognizes the voice input by the input module and establishes a context model; meanwhile, a keyword recognition module recognizes keywords in the context model, a relation inference module outputs inference results according to the keywords, the inference results are matched with a knowledge base storage module, and a relation network model is established. According to the invention, personalized answering for individual users is realized, and in the answering process, autonomous learning and expansion are provided, and the accuracy of the output result is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent answering, and in particular relates to an Internet-based virtual human intelligent answering system. Background Art

[0002] The Internet-based virtual human intelligent answering system is an interactive platform that applies artificial intelligence technology, especially natural language processing (NLP) and text generation technology. It can recognize and analyze the voice or text information input by the user, generate corresponding reply content, and provide real-time feedback to the user through the virtual character image, thereby realizing and enhancing the interactive experience.

[0003] In addition, the construction of the virtual human intelligent answering system also involves a variety of technologies such as deep learning, speech synthesis, sentence analysis, etc. These technologies together create a comprehensive product with human characteristics, which can exist in the non-physical world and appear in various forms, such as virtual assistants, virtual customer service, virtual idols / anchors, etc., and also provide new application scenarios and experiences in the fields of social interaction, entertainment, education, etc., and are also used in emerging scenarios such as smart phone terminals, car-machine interaction, and companion robots.

[0004] Problems with existing technologies:

[0005] Existing Internet-based virtual human intelligent answering systems mostly adopt the question-and-answer method, and cannot achieve relevance in a multi-language, multi-sentence input environment. In the answering process, the relevance to the question and answer varies greatly, and the answers tend to be popular output results. For interactions between individual users, they are not personalized enough and the results are too rigid. Summary of the invention

[0006] The purpose of the present invention is to provide an Internet-based virtual human intelligent answering system that can provide personalized answers to individual users and provide autonomous learning and expansion during the answering process to improve the accuracy of the output results.

[0007] The technical solution adopted by the present invention is as follows:

[0008] A virtual human intelligent answering system based on the Internet, including a virtual human terminal for intelligent voice and image interaction;

[0009] The virtual human terminal includes a power supply module, an input module, a processing unit and an output module;

[0010] The processing unit is provided with a CPU and a speech recognition module, an answer generation module, a relationship inference module, a keyword recognition module, a knowledge base storage module and a domain expansion module connected to the CPU by electrical signals;

[0011] The speech recognition module further recognizes the speech input by the input module and establishes a context model;

[0012] At the same time, the keyword recognition module recognizes the keywords in the context model, and the relationship inference module outputs the inference results based on the keywords, matches the knowledge base storage module, and establishes a relationship network model;

[0013] The answer generation module associates the search results with the context model, the relationship network model and the domain expansion module to make a response, which is then output by the output module.

[0014] Establishing the context model includes the following steps:

[0015] Step 1: The input module inputs the voice into the voice recognition module;

[0016] Step 2: Generate text through the voice recognition module and input it;

[0017] Step 3: First, extract and identify the keywords in a single sentence;

[0018] Step 4: extract and identify keywords associated with the sentence in step 3;

[0019] Step 5: Associating the keywords in step 3 with those in step 4 to determine the conversation semantics and conversation scenario;

[0020] Step 6: Establish a context based on step 5 and generate a context model.

[0021] Establishing the relationship network model includes the following steps:

[0022] Step 7: When generating the context model, the keyword recognition module accesses the context model to a knowledge base storage module related to the keyword;

[0023] Step 8: The relationship inference module and the output result of step 7 jointly generate a relationship network model;

[0024] Step 9: The relational network model after matching the knowledge base storage module transmits the inference results to the answer generation module for making answers, or searches for matching domain expansion modules through the Internet to increase the accuracy and breadth of the answers;

[0025] Step 10: Output the answer result through voice, graphics or gesture through the output module on the virtual human terminal.

[0026] The relationship network model includes group A components, kinship network, geographical network, industry network, encounter network and distant source network, group B components, major event network, medium-sized event network, individual event network and small event network;

[0027] The relationship network model combination method includes the combination method of any components in group A and group B.

[0028] The knowledge base storage module includes a local storage module and a cloud storage module.

[0029] Step 8 Generate a relationship network model. The specific steps are as follows:

[0030] S1, query indexing the keywords through the keyword recognition module in step 7 for input processing;

[0031] S2, matching the input content with the knowledge base storage module, extracting the corresponding content to generate an information set;

[0032] S3, the relational network model constructs a network model based on the relational network model for the received information set;

[0033] S4, extracting the weight value in the relationship network model to reflect the confidence and strength;

[0034] S5, extract useful information, filter and remove redundant information in the knowledge base storage module;

[0035] S6, identify useful information, ensure that the context-related information is consistent, under the same context model, and achieve context fusion;

[0036] S7. Output and generate a relational network model.

[0037] In generating the context model and the relationship network model, a feedback module is provided for receiving interruption information and answering result and adopting system feedback.

[0038] The interruption information includes an interruption instruction during the result output process.

[0039] The answer result adoption system includes the association between the next voice input and the answer result after the virtual human terminal generates the answer, and is used to generate result feedback information after the virtual human terminal generates the answer.

[0040] The technical effects achieved by the invention are:

[0041] The present invention, by establishing a context model and a relationship network model, is used to improve the virtual human terminal's understanding of the dialogue environment and the dialogue network. The context model can set a range for the virtual human terminal's answer content to match questions and answers, thereby improving the accuracy of the output results. By matching the local storage module, the virtual human terminal's answer is made more personalized.

[0042] The present invention provides a thinking and judgment basis path for the virtual human terminal through the established relationship network model, so that the virtual human terminal can index relevant content according to the knowledge base storage module and the domain expansion module, and establish the relevant characters or individuals in the interaction process in the knowledge base storage module, so as to facilitate retrieval and use during the next voice interaction.

[0043] The present invention, through the establishment of a relationship network model, can dynamically adjust the weight of the relationship network model according to the user's question and the context information provided, so that the answer is more in line with the current conversation, and enhances personalization by continuously learning the habits of individual users. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the structure of the present invention;

[0045] Figure 2 It is a structural schematic diagram of the processing unit in the present invention;

[0046] Figure 3 It is a schematic diagram of the steps of establishing a context model in the present invention;

[0047] Figure 4 It is a schematic diagram of the steps of establishing a relationship network model in the present invention;

[0048] Figure 5 It is a structural diagram of the relationship network model in the present invention;

[0049] Figure 6 This is a diagram of a dialogue between a virtual human terminal in the present invention. Figure 1 ;

[0050] Figure 7 This is a diagram of a dialogue between a virtual human terminal in the present invention. Figure 2 ;

[0051] Figure 8 It is a working schematic diagram of the feedback module in the present invention;

[0052] Fig. 9 It is a schematic diagram of the specific steps of generating a relationship network model in the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose and advantages of the present invention more clearly understood, the present invention is specifically described below in conjunction with embodiments. It should be understood that the following text is only used to describe one or several specific embodiments of the present invention, and does not strictly limit the scope of protection of the specific claims of the present invention.

[0054] like Figure 1-4As shown, a virtual human intelligent answering system based on the Internet includes a virtual human terminal for intelligent voice and image interaction. The virtual human terminal includes a power supply module, an input module, a processing unit and an output module.

[0055] Furthermore, the processing unit is provided with a CPU and a speech recognition module, an answer generation module, a relationship inference module, a keyword recognition module, a knowledge base storage module and a domain expansion module connected to the CPU by electrical signals.

[0056] Among them, the speech recognition module further recognizes the speech recorded by the input module and establishes a context model. The specific steps are as follows:

[0057] Step 1: The input module inputs the voice into the voice recognition module;

[0058] Step 2: Generate text through the voice recognition module and input it;

[0059] Step 3: First, extract and identify the keywords in a single sentence;

[0060] Step 4: extract and identify keywords associated with the sentence in step 3;

[0061] Step 5: Associating the keywords in step 3 with those in step 4 to determine the conversation semantics and conversation scenario;

[0062] Step 6: Establish a context based on step 5 and generate a context model.

[0063] At the same time, the keyword recognition module identifies the keywords in the context model, and the relationship inference module outputs the inference results based on the keywords, matches the knowledge base storage module, and establishes a relationship network model. The specific steps are as follows:

[0064] Step 7: When generating the context model, the keyword recognition module accesses the context model to a knowledge base storage module related to the keyword;

[0065] Step 8: The relationship inference module and the output result of step 7 jointly generate a relationship network model;

[0066] Step 9: The relational network model after matching the knowledge base storage module transmits the inference results to the answer generation module for making answers, or searches for matching domain expansion modules through the Internet to increase the accuracy and breadth of the answers;

[0067] Step 10: Output the answer result through voice, graphics or gesture through the output module on the virtual human terminal.

[0068] According to the above, the core calculation of the above-mentioned generated content is realized by the CPU, and the calculated results are distributed and outputted through the connection with the CPU. The voice recognition module, the answer generation module, the relationship inference module, the keyword recognition module, the knowledge base storage module, the domain expansion module and the CPU are bridged. By loading them on the PCB board, the input and output efficiency of the calculation results is realized. The supported protocols include but are not limited to the PCI bus, QPI bus, south bridge and north bridge. Through hardware expansion, it is used to enhance the computing power of the CPU. The above-mentioned CPU needs to have multi-core and multi-threading, high-speed cache and high memory, and high performance and stability.

[0069] Please refer to Fig. 9 As shown, the specific steps for generating the relationship network model in step eight are as follows:

[0070] S1, query indexing the keywords through the keyword recognition module in step 7 for input processing;

[0071] S2, matching the input content with the knowledge base storage module, extracting the corresponding content to generate an information set;

[0072] S3, the relational network model constructs a network model based on the relational network model for the received information set;

[0073] S4, extracting the weight value in the relationship network model to reflect the confidence and strength;

[0074] S5, extract useful information, filter and remove redundant information in the knowledge base storage module;

[0075] S6, identify useful information, ensure that the context-related information is consistent, under the same context model, and achieve context fusion;

[0076] S7. Output and generate a relational network model.

[0077] Furthermore, the principle is:

[0078] By establishing a context model and a relationship network model, the virtual human terminal can improve its understanding of the dialogue environment and dialogue network. The context model can be used to set a range for the answer content of the virtual human terminal to match questions and answers and improve the accuracy of the output results, that is, to tell the virtual human terminal what we are going to talk about;

[0079] By establishing a relationship network model, a path for thinking and judgment is provided for the virtual human terminal, that is, telling the virtual human what the content of the conversation is, so that the virtual human terminal can index the relevant content according to the knowledge base storage module and the domain expansion module, and establish the relevant characters or individuals in the interaction process in the knowledge base storage module, so as to facilitate retrieval and use during the next voice interaction.

[0080] After establishing the context model and the relationship network model, the knowledge base storage module is matched before output to improve the targeted answers to individuals, the field expansion module is used to improve the breadth of knowledge of the virtual human terminal, and the answer generation module is used to make responses based on the context model and output by the output module.

[0081] In addition, when establishing a local storage module, the relationship network information can be entered by voice input or in a conversation such as: So-and-so is my family member, friend, etc., and stored in the local storage module, or by reading the address book.

[0082] The knowledge base storage module includes a local storage module and a cloud storage module.

[0083] Please refer to Figure 5-7 As shown, based on the relationship network stored or recorded in the local storage module, a corresponding relationship network model is established for the relationships in the relationship network, including:

[0084] Group A consists of kinship network, geographical network, professional network, encounter network and distant source network, represented by ka, kb, kc, kd and ke respectively, and the intimacy and weight level are ka>kb>kc>kd>ke;

[0085] Group B consists of major event networks, medium-sized event networks, individual event networks, and small event networks, which are represented by m4, m3, m2, and m1 respectively, and their weight levels are m4>m3>m2>m1;

[0086] Among them, the necessary condition is: ke*m maximum value < kd*m minimum value.

[0087] The context model function formula is:

[0088] Any of context model = f(k, m, k), context model = f(k, m), or context model = f(k, k).

[0089] For example, the types of the relationship network model are set to include w(k, m, k), w(k, m), and w(k, k).

[0090] For example, w(a, 4, b) represents the relationship type between the kinship network and the major event network and the geographical relationship network in the relationship network model.

[0091] According to the above, by combining different types of group A relationship networks with group B event networks, the system can better understand the context of the conversation, which helps the system to more accurately grasp the user's intentions and needs. In addition, through the mutual connection between group A and group B components, the relevance and accuracy of the answer results can be improved. By cooperating with and reading the local storage module, the relationship network model can capture the user's personal preferences and specific social relationships, so that the system can provide more personalized answers, and can also output enhanced logical answers through the relationship network model.

[0092] In addition, as the conversation progresses, the system can dynamically adjust the weights of the relationship network model based on the user's questions and the contextual information provided, so that the answer is more in line with the current state of the conversation, and enhance personalization by constantly learning the habits of individual users. Furthermore, by combining different relationship networks, the system can perform more complex reasoning tasks, such as understanding complex social relationships or the cause and effect of events.

[0093] Furthermore, by calculating the weights of various relationship network models, the virtual human terminal can arrange the information with higher weights during the answering process, extract and classify the information in the sentence or paragraph according to the keyword recognition module, and the timestamp is the default item.

[0094] like Figure 6 and Figure 7 As shown, the virtual human terminal can give corresponding answers based on the split instructions, and provide personalized intelligent answers through the knowledge base data stored locally, combined with indexed cloud data and field expansion data.

[0095] Among them, the timestamp is a deterministic factor, and the machine language is used for execution. As for the factors that affect the results, the solutions are as follows:

[0096] Please refer to Figure 6-8 As shown, in the generation of context model and relationship network model, a feedback module is provided for receiving interruption information and feedback from the answer result adoption system, wherein the interruption information includes interruption instructions in the process of outputting results, wherein the interruption instructions are in the process of the virtual human terminal outputting results, such as running the voice input module at the same time, for receiving specific instructions such as "stop", "pause", "no need", etc., so that the virtual human terminal stops running, or pauses running, and waits for the next instruction input;

[0097] Furthermore, the interruption instruction in the process of outputting the result is used to record whether the output result is wrong, or the output result is not satisfactory, or the approximate output result has been understood and there is no need to continue the output.

[0098] Furthermore, the answer result adoption system includes the correlation between the next voice input and the answer result after the virtual human terminal generates the answer, which is used to generate result feedback information after the virtual human terminal generates the answer.

[0099] Among them, the feedback instruction input after the result is output is used to determine the intention of the interruption instruction. In addition, different from the interruption instruction, the feedback instruction of the next sentence is used to identify together with the question instruction of answering. By combining the context model and the relationship network model, it is used to evaluate the satisfaction of the output result. The evaluation criteria are the correlation between the input instruction, such as the frequency of repetition and the use of modal particles in the next sentence instruction.

[0100] Feedback information is used to rebuild the context model and relationship network model to make them more personalized and match the output results.

[0101] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. A virtual human intelligent answering system based on the Internet, characterized in that: include: Virtual human terminal, used for intelligent voice and image interaction; The virtual human terminal includes a power supply module, an input module, a processing unit and an output module; The processing unit is provided with a CPU and a speech recognition module, an answer generation module, a relationship inference module, a keyword recognition module, a knowledge base storage module and a domain expansion module connected to the CPU by electrical signals; The speech recognition module further recognizes the speech input by the input module and establishes a context model; At the same time, the keyword recognition module recognizes the keywords in the context model, and the relationship inference module outputs the inference results based on the keywords, matches the knowledge base storage module, and establishes a relationship network model; The answer generation module associates the search results according to the context model, the relationship network model and the domain expansion module to make personalized responses to the questions, and generates logical answers according to the relationship network model, which are output through the output module.

2. The Internet-based virtual human intelligent answering system according to claim 1, characterized in that: Establishing the context model includes the following steps: Step 1: The input module inputs the voice into the voice recognition module; Step 2: Generate text through the voice recognition module and input it; Step 3: First, extract and identify the keywords in a single sentence; Step 4: extract and identify keywords associated with the sentence in step 3; Step 5: Associating the keywords in step 3 with those in step 4 to determine the conversation semantics and conversation scenario; Step 6: Establish a context based on step 5 and generate a context model.

3. The Internet-based virtual human intelligent answering system according to claim 1, characterized in that: Establishing the relationship network model includes the following steps: Step 7: When generating the context model, the keyword recognition module accesses the context model to a knowledge base storage module related to the keyword; Step 8: The relationship inference module and the output result of step 7 jointly generate a relationship network model; Step 9: The relational network model after matching the knowledge base storage module transmits the inference results to the answer generation module for making answers, or searches for matching domain expansion modules through the Internet to increase the accuracy and breadth of the answers; Step 10: Output the answer result through voice, graphics or gesture through the output module on the virtual human terminal.

4. The Internet-based virtual human intelligent answering system according to claim 1, characterized in that: The relationship network model includes group A components, kinship network, geographical network, industry network, encounter network and distant source network, group B components, major event network, medium-sized event network, individual event network and small event network; The relationship network model combination method includes the combination method of any components in group A and group B.

5. The Internet-based virtual human intelligent answering system according to claim 3 is characterized by: The knowledge base storage module includes a local storage module and a cloud storage module.

6. The Internet-based virtual human intelligent answering system according to claim 1, characterized in that: Step 8 Generate a relationship network model. The specific steps are as follows: S1, query indexing the keywords through the keyword recognition module in step 7 for input processing; S2, matching the input content with the knowledge base storage module, extracting the corresponding content to generate an information set; S3, the relational network model constructs a network model based on the relational network model for the received information set; S4, extracting the weight value in the relationship network model to reflect the confidence and strength; S5, extract useful information, filter and remove redundant information in the knowledge base storage module; S6, identify useful information, ensure that the context-related information is consistent, under the same context model, and achieve context fusion; S7. Output and generate a relational network model.

7. The Internet-based virtual human intelligent answering system according to claim 6, characterized in that: In generating the context model and the relationship network model, a feedback module is provided for receiving interruption information and answering result and adopting system feedback.

8. The Internet-based virtual human intelligent answering system according to claim 7, characterized in that: The interruption information includes an interruption instruction during the result output process.

9. The Internet-based virtual human intelligent answering system according to claim 7, characterized in that: The answer result adoption system includes the association between the next voice input and the answer result after the virtual human terminal generates the answer, and is used to generate result feedback information after the virtual human terminal generates the answer.

Images