Artificial Intelligence-Based Intelligent Robot Q&A Method and System

Abstract

The present invention provides an intelligent robot question-answering method and system based on artificial intelligence. The method includes the following steps: collecting an audio signal of a user's question and preprocessing the audio signal; dividing the preprocessed audio signal into multiple speech signal segments; identifying keyword entities in each of the speech signal segments and calculating the importance degree of each of the keyword entities; determining the analysis priority of each of the speech signal segments according to the importance degree of the keyword entities; and generating an optimized reply to the user's question based on the analysis priority, thereby reducing the phenomenon that the intelligent robot customer service answers the user's question irrelevantly and improving the user experience.

Description

Technical Field

[0001] This application relates to the field of data processing, and more particularly to an intelligent robot question-answering method and system based on artificial intelligence. Background Art

[0002] With the development of deep learning (such as the Transformer architecture) and pre-trained language models (such as GPT and BERT), machines can more accurately understand the semantics, context, and even emotions of human language, enabling intelligent question-answering systems to upgrade from "keyword matching" to "semantic understanding". And with the continuous penetration of intelligent robots in the consumer market, intelligent robots have been applied in various industries, such as intelligent customer service. However, there are still some problems with intelligent robot customer service, which can easily lead to a poor user experience.

[0003] Intelligent robot customer service mainly extracts information from its knowledge base by identifying keyword information in the user's question. However, when the user's question is relatively complex, that is, it contains a large amount of keyword information, the intelligent robot customer service may not be able to accurately determine the core of the question raised by the user, easily causing misunderstandings or answering off-topic, resulting in a poor user experience. Summary of the Invention

[0004] In order to reduce the phenomenon of intelligent robot customer service answering off-topic to user questions and improve the user experience, this application provides an intelligent robot question-answering method and system based on artificial intelligence.

[0005] In a first aspect, this application provides an intelligent robot question-answering method based on artificial intelligence, adopting the following technical solutions:

[0006] Collect the audio signal of the user's question and preprocess the audio signal;

[0007] Divide the preprocessed audio signal into multiple speech signal segments;

[0008] Identify keyword entities in each of the speech signal segments and calculate the importance of each keyword entity;

[0009] Determine the analysis priority of each speech signal segment according to the importance of the keyword entity;

[0010] Generate an optimized reply to the user's question based on the analysis priority;

[0011] Among them, calculating the importance of each keyword entity includes:

[0012] Extract the syntactic relationships between each of the speech signal segments, and construct a local relationship graph with each of the keyword entities as nodes and the dependency relationships between each of the keyword entities as edges;

[0013] Calculate the domain-frequency weights of the keyword entities;

[0014] For keyword entities with dependency relationships, calculate the cumulative value of the path lengths between the keyword entity and other keyword entities on the local relationship graph;

[0015] Take the ratio of the total number of all keyword entities to the cumulative value of the path lengths of the keyword entities with dependency relationships, and the product of the total number of keyword entities and the domain-frequency weights of the keyword entities as the importance degree of the keyword entities.

[0016] Through specific analysis of the speech signal of the user's question, the present invention divides the entire speech signal segment, and infers the analysis priorities of different segments by combining the importance degrees of different keyword entities. By analyzing the analysis priorities of different segments and combining the importance degrees of different keyword entities, this method can enable the intelligent robot customer service to accurately find the core concerns of the user's question, generate an initial response based on this, and further optimize the initial response by combining other segments, reducing the phenomenon that the intelligent robot customer service answers the user's question irrelevantly and improving the user experience.

[0017] Optionally, the preprocessing includes:

[0018] Perform noise reduction processing on the collected audio signal, including eliminating power frequency interference and non-stationary noise;

[0019] Perform acoustic echo cancellation on the audio signal to remove the interference of the device's own playback sound.

[0020] Use the voice collection module embedded in the intelligent robot to collect the audio information representing the user's question, process the collected audio information, reduce the noise influence, and enhance the performance of the user's audio signal.

[0021] Optionally, the dividing the preprocessed audio signal into multiple speech signal segments includes:

[0022] Divide the audio signal into multiple speech signal segments based on the interruption characteristics of the speech signal and the time performance of adjacent speech signals.

[0023] Based on the collected audio signal representing the user's question, divide the overall collected user audio signal into segments to accurately identify the keyword entities in different segments.

[0024] Optionally, dividing the audio signal into multiple speech signal segments based on the interruption feature of the speech signal and the temporal representation of adjacent speech signals includes:

[0025] Mark the discontinuous parts on the entire speech signal segment along the starting point of the speech signal;

[0026] Calculate the corresponding interruption possibility at the discontinuous part;

[0027] If the interruption possibility conforms to the interruption feature and the duration of the undivided speech signal before the discontinuous part is greater than the set time threshold, divide at the discontinuous part and obtain corresponding multiple speech signal segments.

[0028] Dividing the entire speech signal segment through the interruption feature of the user speech signal and the temporal representation of adjacent speech signals facilitates accurately finding more core entity keywords from multiple speech signal segments.

[0029] It also includes: If the sum of the duration of the undivided speech signal before the discontinuous part and the duration of the first continuous speech signal after the discontinuous part is greater than the set time threshold, divide at the discontinuous part and obtain corresponding multiple speech signal segments;

[0030] If the sum of the duration of the divided speech signal before the discontinuous part and the duration of the first continuous speech signal after the discontinuous part is less than or equal to the set time threshold, reduce the division weight corresponding to the discontinuous part.

[0031] By appropriately reducing the division weight corresponding to the discontinuous part, the accuracy of speech signal segment division is further improved.

[0032] Optionally, identifying the keyword entities in each of the speech signal segments includes:

[0033] Convert the speech signals of different segments into acoustic feature sequences, combine with context information to convert them into text, and use natural language processing technology to extract keyword entities from the recognized text.

[0034] By converting the speech signals of different segments into acoustic feature sequences and converting them into corresponding text information, it can further help the intelligent robot customer service quickly and accurately identify the key information in the user's words.

[0035] Optionally, calculating the importance degree of the keyword entities includes:

[0036] Based on the hierarchical depth, domain coefficient, and occurrence frequency of the keyword entity in the knowledge graph, calculate the domain-frequency weight of the keyword entity;

[0037] Determine the importance of the keyword entity based on the domain-frequency weight.

[0038] By assigning different weight values to different entity keywords, calculate the importance of different keywords in the user's question, so as to extract the core point of the user's question, and then help the intelligent robot to give the user the correct answer.

[0039] Optionally, the determination of the analysis priority of the speech signal segment includes:

[0040] Calculate the average value of the importance of the keyword entities in each speech signal segment;

[0041] Determine the analysis priority of each speech signal segment according to the average value of the importance of the keyword entities.

[0042] Based on the analysis priority of the text information corresponding to different segments of speech signals, help the intelligent robot extract the key information of the user's question, so as to generate a more accurate question answer.

[0043] Optionally, the generation of the optimized reply includes:

[0044] Based on the analysis priority of the text information corresponding to different segments of speech signals obtained, construct a data sequence from largest to smallest;

[0045] Based on the important keyword entities in the speech signal segment with the highest analysis priority, identify the core intention of the user's question and generate a basic reply to the question mentioned by the user;

[0046] Add the keyword entities of other speech signal segments in the data sequence in turn according to the analysis priority, optimize the reply content and generate a question answer pair to reply to the user accordingly.

[0047] Add the keyword entities of other speech signal segments in the data sequence in turn according to the analysis priority to optimize the reply content, so as to generate a reply that more conforms to the user's question.

[0048] In a second aspect, the present application provides an intelligent robot question-answering system based on artificial intelligence, adopting the following technical solution:

[0049] A processor and a memory, the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the intelligent robot question-answering method based on artificial intelligence according to any one of claims 1-9 is implemented.

[0050] Generate a computer program for the above-mentioned intelligent robot question-answering method based on artificial intelligence and store it in the memory to be loaded and executed by the processor. Thus, a system is made according to the memory and the processor, which is convenient to use.

[0051] The present application has the following technical effects:

[0052] By specifically analyzing the voice signal of the user's question, the present invention divides the entire voice signal segment, and infers the analysis priority of different segments by combining the importance of different keyword entities. By analyzing the analysis priority of different segments and combining the importance of different keyword entities, this method can enable the intelligent robot customer service to accurately find the core concerns of the user's question, generate an initial reply based on this, and further optimize the initial reply by combining other segments, reducing the phenomenon that the intelligent robot customer service answers off-topic to the user's question and improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a flowchart of the method of steps S1 - S5 in the intelligent robot question - answering method based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] An embodiment of the present application discloses an intelligent robot question - answering method based on artificial intelligence. Referring to Figure 1 , it includes:

[0055] S1: Collect the audio signal of the user's question and pre - process the audio signal.

[0056] In one embodiment, the voice acquisition module embedded in the intelligent robot is used to collect the audio information representing the user's question, and the collected audio information is processed to reduce the influence of noise and enhance the performance of the user's audio signal. Among them, the specific steps of the pre - processing are as follows:

[0057] S11: Perform noise reduction processing on the collected audio signal, including eliminating power frequency interference and non - stationary noise;

[0058] S12: Perform acoustic echo cancellation on the audio signal to remove the interference of the device's own playback sound.

[0059] As an example, the voice acquisition module embedded in the intelligent robot can be used to collect the user's question audio signal. Set the sampling rate to 16 kHz; the bit depth to 16 bit; the storage format to PCM lossless storage. At the same time, an acoustic echo cancellation (AEC) module is built - in to remove the interference of the device's own playback sound. When the intelligent robot does not receive the user's question voice signal for 5 consecutive seconds, the voice acquisition module stops working, indicating that the current audio signal acquisition has been completed.

[0060] Optionally, a Wiener filter is used to eliminate 50Hz power frequency interference, and spectral subtraction is combined to suppress non - stationary noise, completing the pre - processing of the collected audio signal representing the user's question.

[0061] S2: Divide the preprocessed audio signal into multiple speech signal segments.

[0062] It should be noted that when a user asks a question to the intelligent robot customer service, the text information contained in the question may be excessive, resulting in the intelligent robot customer service being unable to accurately extract the user's core concerns. The answer given to the user may deviate from what the user wants, resulting in an off-topic answer and affecting the user experience. Therefore, based on the collected audio signal representing the user's question, this application divides the overall collected user audio signal into segments and identifies the keyword entities in different segments. Among them, dividing the preprocessed audio signal into multiple speech signal segments includes:

[0063] S21: Based on the interruption characteristics of the speech signal and the time performance of adjacent speech signals, divide the audio signal into multiple speech signal segments;

[0064] It should be noted that when the language of the user's question to the intelligent robot is long, directly extracting entity keywords from the speech signal may cause the intelligent robot to misjudge the key information because of the relatively large number of consecutive entity keywords. As a result, the intelligent robot may have difficulty correctly understanding the correct meaning of the user's question, resulting in an off-topic answer and affecting the user experience. Therefore, this application analyzes and processes the preprocessed user speech signal collected in the above steps to help the intelligent robot understand the question raised by the user.

[0065] Specifically, obtain the preprocessed user's language audio signal for asking questions. The speech audio signal here is the complete audio signal collected from the user, which may be a piece of language text or multiple consecutive pieces of language text with an indefinite length. A long speech signal may contain a large number of entity keywords, and it is difficult for the intelligent robot to find the core among so many entity keywords. Therefore, this application analyzes the collected speech signal representing the user's question and divides a long speech signal into multiple small segments. Specifically as follows:

[0066] The specific analysis of dividing the audio signal into multiple speech signal segments includes:

[0067] S210: Starting from the starting point of the speech signal, mark the discontinuous parts on the entire speech signal segment;

[0068] It should be noted that the discontinuous parts here are not a single point, but multiple consecutive points.

[0069] S211: Calculate the corresponding interruption possibility at the discontinuous parts;

[0070] In the embodiment of this application, the discontinuous part is the x-th position in the speech signal. Calculate the Possibility of user's ventilation or short interruption at the position:

[0071]

[0072] In the formula, is the possibility of user's ventilation or short interruption at the th position (corresponding to the discontinuous part) in the voice signal. represents the duration of the th position, that is, the time of short interruption of the voice signal at this position. represents the average intensity in the voice signal before the th position. represents the signal intensity of the last peak in the voice signal before the th position. represents the duration from the last peak to the valley in the voice signal before the th position.

[0073] It should be noted that the above formula is mainly analyzed and calculated based on the duration of the th position and the signal performance before the interruption of the voice signal. Generally, when people ventilate or interrupt while speaking, they will choose a more appropriate place in the language to interrupt. When the voice signal is briefly interrupted in a certain part, and the intensity of the voice signal decreases and drops rapidly before the interruption, it indicates that the user has made a short ventilation or a short interruption of speaking here.

[0074] S212: If the interruption possibility conforms to the interruption characteristics and the duration of the unpartitioned voice signal before the discontinuous part is greater than the set time threshold, then partition at the discontinuous part and obtain corresponding multiple voice signal segments.

[0075] In the embodiments of the present application, the interruption characteristics are as follows: Generally, when people ventilate or interrupt while speaking, they will choose a more appropriate place in the language to interrupt. When the voice signal is briefly interrupted in a certain part, and the intensity of the voice signal decreases and drops rapidly before the interruption, it indicates that the user has made a short ventilation or a short interruption of speaking here.

[0076] As an example, the interruption possibility can be judged whether it conforms to the interruption characteristics in the following way: preprocess the voice signal (such as denoising, enhancing the signal); calculate the change of the intensity (energy) of the voice signal over time; detect the points where the intensity of the signal suddenly decreases or drops rapidly; combine the intensity changes before and after the signal interruption to judge whether it conforms to the characteristics of ventilation or short interruption.

[0077] Optionally, the formula can be: Perform voice signal intensity variation analysis. In the formula, E(t) is the energy of the voice signal at time t, θenergy is the threshold of energy variation, Δt is the time window, 1 indicates that this position conforms to the interruption feature (sudden reduction or disappearance of high-frequency components), and 0 indicates that this position does not conform to the interruption feature (large variation rate of high-frequency components or no significant reduction).

[0078] In addition, through spectral feature analysis:

[0079] Perform short-time Fourier transform (STFT) on the voice signal, extract spectral features, detect sudden reduction or disappearance of high-frequency components in the spectrum, and combine spectral changes and time features to determine whether it conforms to the features of breathing or interruption.

[0080] Among them, the formula is: , where HighFreq(t) is the high-frequency component of the voice signal at time t, and θspectral is the threshold of spectral variation. If the variation rate of the high-frequency component of the voice signal at time t (i.e., ) is not less than the threshold θspectral, the return value is 0, and otherwise is used to handle the situation that does not conform to the interruption feature, that is, when the variation rate of the high-frequency component of the voice signal does not conform to the interruption feature, the return value is 0.

[0081] Furthermore, combine the above voice signal intensity variation and spectral features to construct a comprehensive interruption possibility score:

[0082]

[0083] In the formula, α and β are weight coefficients used to balance the influence of intensity variation and spectral features, and Pcombined(t) is the comprehensive interruption possibility score.

[0084] In an embodiment of the present application, calculate the intensity variation probability Pinterruption(t) and spectral feature probability Pspectral(t) of the voice signal at a certain time point t. Among them, Pinterruption(t) = 0.7, Pspectral(t) = 0.4, and then determine the comprehensive interruption possibility score Pcombined(t) of the above two probabilities. To balance the influence of the above two features (intensity variation probability and spectral feature probability), select α = 0.6 and β = 0.4. Calculate the comprehensive interruption possibility score: Pcombined(t) = 0.42 + 0.16 = 0.58.

[0085] As an example, if Pcombined(t) exceeds a set threshold, it is determined that this position is a discontinuous part of the user's speech, conforming to the characteristics of breathing or a short interruption. A signal processing library (such as librosa or numpy in Python) can be used to implement the above analysis. Through the above analysis of the interruption characteristics, it can be objectively and accurately determined whether the interruption possibility at the discontinuous part of the user's speech signal conforms to the characteristics of breathing or a short interruption. For example, the set threshold is 0.5 and Pcombined(t) = 0.58, then at this time the discontinuous part of the user's speech conforms to the characteristics of breathing or a short interruption.

[0086] Optionally, when performing the division in this application, the time performance of the user's question is also considered. By setting a duration threshold , to help further calculate the possibility of division at the position in the speech signal:

[0087]

[0088] In the formula, is the possibility of division at the position (corresponding to the discontinuous part) in the speech signal, represents the duration of the continuous speech signal that has not been divided before the position in the speech signal, is the standard normalization function, represents the influence performance of the division at the x position by the duration of the first continuous speech signal after this position. When , ; when , 0.5. represents the duration of the first continuous speech signal after the position in the speech signal.

[0089] It should be noted that the above formula is mainly calculated based on the possibility that the position is a user's breathing or short interruption, combined with the duration of the surrounding continuous speech signals. When the possibility that the position is a user's breathing or short interruption is greater, and the duration of the speech signal before the position that has been divided is closer to or greater than the threshold duration, then the probability of division at this position is greater.

[0090] As an example, set the threshold , when the calculated , it means that division can be performed at the position in the speech signal.

[0091] In addition, the present application also considers the duration of the first continuous speech signal after the position in the speech signal. If the sum of the two times is close to or greater than the set time threshold, division can be performed here; if the sum of the two times is less than or equal to the set time threshold, the weight of division here is appropriately reduced.

[0092] Repeat the above steps to divide the collected speech signal of the user's question into multiple segments of speech signals.

[0093] S3: Identify the keyword entities in each speech signal segment and calculate the importance degree of each keyword entity.

[0094] It should be noted that in different speech signals, the text information carried by them is different, and the importance of different texts to the entire discourse is also different. Based on the above-mentioned multiple segments of speech signals divided, the present application identifies the keywords in different speech signal segments, further analyzes the importance degree of different keywords in the user's question discourse, extracts the core points of the user's question, and then helps the intelligent robot to feedback the correct answer to the user.

[0095] Identifying the keyword entities in each speech signal segment includes:

[0096] S31: Convert the speech signals of different segments into acoustic feature sequences, and convert them into text in combination with context information. Using natural language processing technology, extract the keyword entities from the identified text.

[0097] Optionally, calculating the importance degree of the keyword entities includes:

[0098] S311: Calculate the domain-frequency weight of the keyword based on the hierarchical depth, domain coefficient, and occurrence frequency of the keyword entity in the knowledge graph;

[0099] S312: Determine the importance degree of the keyword entity based on the domain-frequency weight.

[0100] In the embodiment of the present application, the hierarchical depth of the th keyword entity in the knowledge graph is obtained based on the knowledge graph in the intelligent robot knowledge base , and the domain-frequency weight of the th keyword is calculated :

[0101]

[0102]

[0103] In the formula, represents the Frequency performance of keyword entities Indicates the number of times the keyword appears; Indicates the average number of times all keyword entities appear in the user's question. Indicates the number of segments containing the keyword. Indicates the maximum hierarchical depth in the knowledge graph; Indicates the domain coefficient of the domain corresponding to the keyword entity (initially set in the system).

[0104] It should be noted that the above formula is mainly calculated based on the hierarchical depth of the keyword in the knowledge graph, the domain coefficient, and the frequency performance of the keyword. The deeper the hierarchical depth of a keyword in the knowledge graph, the more specific the keyword entity mentioned by the user. Further combined with the size of the domain coefficient of the domain corresponding to the keyword entity (the size of the domain coefficient has a certain calibration significance), if the domain coefficient is also large, it indicates that the keyword entity is more important in the question mentioned by the user. At the same time, when the frequency of a keyword entity in the divided segments is larger, it can also reflect the importance of the keyword entity.

[0105] As an example, calculating the importance of each keyword entity also includes:

[0106] Extracting the syntactic relationships between each speech signal segment, using each keyword entity as a node and the dependency relationships between each keyword entity as edges to construct a local relationship graph; calculating the domain-frequency weight of the keyword entity; for keyword entities with dependency relationships, calculating the cumulative value of the path lengths of the keyword entity and other keyword entities on the local relationship graph; taking the ratio of the total number of all keyword entities to the cumulative value of the path lengths of the keyword entities with dependency relationships and the product of the total number of keyword entities and the domain-frequency weight of the keyword entity as the importance of the keyword entity.

[0107] Specifically, the existing technology can be used to extract the syntactic relationships in each divided segment, using the keyword entity as a node and the dependency relationship as an edge to construct a local relationship graph.

[0108] Further calculate the importance of the keyword entity:

[0109]

[0110] In the formula, is the importance of the keyword entity, represents all segments related to the The number of keyword entities that have a dependency relationship with a keyword entity (including direct and indirect dependencies). Indicates the keyword entity and the other keyword entity's path length on the constructed local relationship graph (only calculate the keyword entities with dependency relationships here).

[0111] It should be noted that the above formula mainly analyzes based on the domain-frequency weight sizes of different keywords, combined with the path length of the dependency relationship between two keyword entities. When the domain-frequency weight of a keyword entity is larger, and the dependency relationship path between this keyword and other keywords is shorter, it indicates that this keyword entity is more important in each segment, that is, the importance degree of the keyword is greater.

[0112] Repeat the above steps to obtain the importance degrees of all keyword entities.

[0113] S4: Determine the analysis priority of each speech signal segment according to the importance degree of the keyword entity.

[0114] Specifically, determining the analysis priority of each speech signal segment includes:

[0115] S41: Calculate the average value of the importance degrees of the keyword entities in each speech signal segment;

[0116] S42: Determine the analysis priority of each speech signal segment according to the average value of the importance degrees of the keyword entities.

[0117]

[0118] In the formula, represents the number of keyword entities recognized in the th segment of the divided speech signal, is the data sequence subtracted from the analysis priority of each speech signal segment, represents the importance degree of the

[0119] It should be noted that the above formula mainly uses the average value of the importance degrees of the keyword entities within the segment to represent the analysis priority of the segment.

[0120] Thus, the analysis priorities of the text information corresponding to different segments of the speech signal are obtained.

[0121] S5: Generate an optimized response to the user's question based on the analysis priority.

[0122] In the embodiments of the present application, the optimized response specifically includes:

[0123] S51: Based on the analysis priorities of the text information corresponding to different segments of speech signals obtained, construct a data sequence from largest to smallest;

[0124] S52: Identify the core intention of the user's question based on the important keyword entities in the speech signal segment with the highest analysis priority, and generate a basic reply to the question mentioned by the user;

[0125] S53: Add the keyword entities of other speech signal segments in the data sequence in turn according to the analysis priorities, optimize the reply content and generate a question-answer pair to reply to the user accordingly.

[0126] S54: Analyze the priorities and generate an optimized reply to the user's question.

[0127] Specifically, the analysis priorities of the text information corresponding to the above different segments of speech signals are obtained. Based on this information, it can help the intelligent robot extract the key information of the user's question, so as to answer the question. Specifically as follows:

[0128] First, based on the analysis priorities of the text information corresponding to the different segments of speech signals obtained above, construct a data sequence from largest to smallest: . The intelligent robot identifies the core intention of the user's question based on the important keyword entities in the speech signal segment with the highest analysis priority, and generates a basic reply to the question mentioned by the user. The remaining speech signal segments are continuously added according to the analysis priorities to optimize the reply to the question mentioned by the user. After the analysis is completed, the intelligent robot replies to the user accordingly based on the generated question-answer pair, and the reply forms include: text + voice.

[0129] The embodiment of the present application also discloses an intelligent robot question-answering system based on artificial intelligence, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the intelligent robot question-answering method based on the present application is implemented.

[0130] The above system also includes other components well known to those skilled in the art such as a communication bus and a communication interface, and their settings and functions are known in the art, so they will not be described in detail here.

[0131] In the present application, the foregoing memory can be any tangible medium that contains or stores a program, and this program can be used or combined with an instruction execution system, device, or device. For example, a computer-readable storage medium can be any suitable magnetic storage medium or magneto-optical storage medium, such as, for example, a resistive random access memory, a dynamic random access memory, a static random access memory, etc., or any other medium that can be used to store the required information and can be accessed by an application program, module, or both. Any such computer storage medium can be part of the device or accessible or connectable to the device.

[0132] The above are all preferred embodiments of the present application, and do not limit the protection scope of the present application accordingly. Therefore, all equivalent changes made according to the structure, shape, and principle of the present application shall be covered within the protection scope of the present application.

Claims

1. An intelligent robot Q&A method based on artificial intelligence, characterized in that, It includes the following steps: Collect the audio signal of the user's question and preprocess the audio signal; Divide the preprocessed audio signal into multiple speech signal segments; Identify the keyword entities in each of the speech signal segments and calculate the importance degree of each keyword entity; Determine the analysis priority of each speech signal segment according to the importance degree of the keyword entity; Generate an optimized reply to the user's question based on the analysis priority; Among them, calculating the importance degree of each keyword entity includes: Extract the syntactic relationship between each speech signal segment, construct a local relationship graph with each keyword entity as a node and the dependency relationship between each keyword entity as an edge; Calculate the domain-frequency weight of the keyword entity; For keyword entities with a dependency relationship, calculate the cumulative value of the path lengths of the keyword entity and other keyword entities on the local relationship graph; Take the ratio of the total number of all keyword entities to the cumulative value of the path lengths of the keyword entities with a dependency relationship, and the product of the total number of keyword entities and the domain-frequency weight of the keyword entity as the importance degree of the keyword entity.

2. The intelligent robot Q&A method based on artificial intelligence according to claim 1, wherein, The preprocessing includes: Perform noise reduction processing on the collected audio signal, including eliminating power frequency interference and non-stationary noise; Perform acoustic echo cancellation on the audio signal to remove the interference of the device's own playback sound.

3. The intelligent robot Q&A method based on artificial intelligence according to claim 1, wherein The dividing the preprocessed audio signal into multiple speech signal segments includes: Based on the interruption characteristics of the speech signal and the time performance of adjacent speech signals, divide the audio signal into multiple speech signal segments.

4. The intelligent robot question-answering method based on artificial intelligence according to claim 3, characterized in that, The dividing the audio signal into multiple speech signal segments based on the interruption characteristics of the speech signal and the time performance of adjacent speech signals includes: Mark the discontinuous parts on the entire speech signal segment along the starting point of the speech signal; Calculate the corresponding interruption possibility at the discontinuous part; If the interruption possibility conforms to the interruption characteristics and the duration of the unsegmented speech signal before the discontinuous part is greater than the set time threshold, then perform segmentation at the discontinuous part and obtain corresponding multiple speech signal segments.

5. The intelligent robot Q&A method based on artificial intelligence according to claim 4, wherein It also includes: If the sum of the duration of the unsegmented speech signal before the discontinuous part and the duration of the first continuous speech signal after the discontinuous part is greater than the set time threshold, then perform segmentation at the discontinuous part and obtain corresponding multiple speech signal segments; If the sum of the duration of the segmented speech signal before the discontinuous part and the duration of the first continuous speech signal after the discontinuous part is less than or equal to the set time threshold, then reduce the segmentation weight corresponding to the discontinuous part.

6. The intelligent robot question-answering method based on artificial intelligence according to claim 1, characterized in that, The identifying the keyword entities in each of the speech signal segments includes: Convert the speech signals of different segments into an acoustic feature sequence, combine the context information and convert it into text, and use natural language processing technology to extract keyword entities from the recognized text.

7. The intelligent robot Q&A method based on artificial intelligence according to claim 1, characterized in that Calculating the importance degree of keyword entities includes: Based on the hierarchical depth, domain coefficient and occurrence frequency of the keyword entity in the knowledge graph, calculate the domain-frequency weight of the keyword entity; Determine the importance level of the keyword entity based on the domain-frequency weight.

8. The intelligent robot Q&A method based on artificial intelligence according to claim 1, wherein The determination of the analysis priority of the speech signal segment includes: Calculate the average value of the importance levels of the keyword entities in each speech signal segment; Determine the analysis priority of each speech signal segment according to the average value of the importance levels of the keyword entities.

9. The intelligent robot question and answer method based on artificial intelligence according to claim 1, characterized in that, Generating an optimized response to the user's question includes: Based on the analysis priorities of the text information corresponding to the different-segment speech signals obtained, construct a data sequence from largest to smallest; Based on the important keyword entities in the speech signal segment with the highest analysis priority, identify the core intention of the user's question and generate a basic response to the question mentioned by the user; Add the keyword entities of the other speech signal segments in the data sequence in turn according to the analysis priority, optimize the response content and generate a question-answer pair to respond to the user accordingly.

10. An intelligent robot Q&A system based on artificial intelligence, characterized in that, Includes: A processor and a memory, the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the artificial intelligence-based intelligent robot question-answering method according to any one of claims 1-9 is implemented.

Images