Personalized intelligent question-answering system based on artificial intelligence

Abstract

The present invention relates to the field of intelligent question-answering technology, and more specifically to a personalized intelligent question-answering system based on artificial intelligence. The system activates a facial photo acquisition system based on question voice and collects facial photos of the current user. It extracts target features representing the face from the facial photos, calculates similarity between the target features and high-dimensional features stored in a question-answering client, and determines the legitimacy of the user based on the similarity. The question voice of the legitimate user is substituted into a speech-to-text recognition model to output a text of the speech content. The system calculates the module similarity between the correlation vector matrix between entity references and the relationship matrix corresponding to the central entity reference in the answer library, and outputs the preset answer text corresponding to the central entity reference with the greatest module similarity. This improves the personalization, accuracy, and intelligence of question-answering, enhances information acquisition efficiency, and enhances the scalability and flexibility of the system.

Description

Technical Field

[0001] The present invention relates to the field of intelligent question-answering technology, and in particular to a personalized intelligent question-answering system based on artificial intelligence. Background Art

[0002] In recent years, artificial intelligence (AI) has provided strong support for the development of intelligent question-answering systems, which leverage natural language processing to understand user questions and quickly provide accurate answers using knowledge graphs and machine learning algorithms.

[0003] However, existing intelligent question-answering systems often suffer from several issues. For one thing, they often lack personalization, failing to provide customized responses based on users' preferences, historical records, and other information. Furthermore, they often lack precision in semantic recognition of user questions, resulting in answers that fail to meet user needs.

[0004] Therefore, there is an urgent need for a personalized intelligent question-answering system based on artificial intelligence to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a personalized intelligent question-answering system based on artificial intelligence: to solve the technical problems that the question-answering system in the prior art cannot provide customized answers based on the user's personal preferences, historical records and other information, and the accuracy and intelligence of the questions and answers are low.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A personalized intelligent question-answering system based on artificial intelligence, which includes:

[0008] The legitimacy judgment module is used to obtain the user's voice question, activate the face photo collection system based on the question voice, collect the current user's face photo, extract the target features representing the face from the face photo, calculate the similarity between the target features and the high-dimensional features stored in the question-answering client, and judge the legitimacy of the user based on the similarity;

[0009] The speech-to-text recognition module is used to substitute the legitimate user's question speech into the speech-to-text recognition model and output the speech content text;

[0010] The entity reference extraction module is used to extract entity references from the speech content text, determine the correlation between entity references in the speech content text based on association rule mining, perform PCA analysis on the associated entity references, and obtain the correlation vector matrix between the associated entity references;

[0011] The answer text output module is used to calculate the module similarity between the correlation vector matrix between entity references and the relationship matrix corresponding to the central entity reference in the answer library, and output the preset answer text corresponding to the central entity reference with the largest module similarity.

[0012] Furthermore, similarity calculation between the target feature and the high-dimensional feature stored in the question-answering client specifically includes the following process:

[0013] Calculate the basic similarity of the feature vector values ​​between the target feature and the high-dimensional feature:

[0014] ;

[0015] in, is the basic similarity, represents the i-th eigenvector value of the target feature representing the face, It represents the i-th feature vector value of the high-dimensional feature vector stored in the Q&A client, n is the dimension of the feature vector, and the high-dimensional features stored in the Q&A client are pre-extracted user facial features.

[0016] Furthermore, judging the legitimacy of a user based on similarity specifically includes the following process:

[0017] Obtain the time of the user's last question and answer and the time of the current question and answer, calculate the time interval between the last question and answer and the current question and answer, and record this time interval as the target time interval; sequentially calculate the historical similarity between all historical time intervals of the user and the query history time interval;

[0018] Number the historical time intervals and the target time interval to obtain numbered values. A rectangular coordinate system is constructed with the numbered values ​​as the X-axis and the historical similarity and similarity as the Y-axis. Adjacent similarities in the rectangular coordinate system are connected to generate a similarity curve. A perpendicular line is drawn from both ends of the similarity curve to the X-axis to obtain two start and end line segments. A closed figure is formed by the similarity curve, the two start and end line segments, and the X-axis. The total area of ​​the closed figure is calculated.

[0019] Determine whether the total area exceeds a preset threshold. If so, the user is determined to be a legitimate user. If not, the user is determined to be an illegal user and is denied access to the question-answering system.

[0020] Furthermore, the legitimate user's question speech is substituted into the speech-to-text recognition model, and the output of the speech content text specifically includes the following process:

[0021] For problem speech data , is a set of problem speech data samples;

[0022] The speech content recognition model consists of two parts: the speech content encoding network f and the speech content context network. The speech content encoder network f consists of a five-layer convolutional network. The speech content encoder outputs a low-frequency feature. , the hidden space after the encoder network mapping is represented by Z, where :

[0023] ;

[0024] Among them, m is the convolution receptive field;

[0025] The speech content context network is a convolutional neural network with the same structure as the speech content encoding network f. It maps the hidden space into the context information space C. The output of the context network is ,in, :

[0026] ;

[0027] Among them, v is the convolution receptive field;

[0028] Loss function based on speech content recognition model To train the speech content recognition model, the loss function It is expressed as follows:

[0029] ;

[0030] in, is the sequence length of the problem speech data sample set X, is the loss of noise contrast estimation, is the true probability of the speech data sample to be evaluated, is the learning parameter;

[0031] Output the speech content text.

[0032] Furthermore, extracting entity references from the speech content text specifically includes: extracting entity references from the speech content text based on the BERT model.

[0033] Furthermore, PCA analysis is performed on the associated entity references to obtain the correlation vector matrix between the associated entity references. The specific process includes the following:

[0034] (1) Form the matrix M of the associated entity references and calculate the covariance matrix S of the samples of the matrix M;

[0035] (2) Calculate the eigenvectors of the covariance matrix S 、 …、 and eigenvalues ​​t = 1,2,…, ;

[0036] (3) Project the data into the feature vector space, using the formula:

[0037] ;

[0038] in, The value is the feature value corresponding to the associated entity reference in the original sample;

[0039] (4) Select R vectors and form a correlation vector matrix.

[0040] Furthermore, the module similarity requirement between the correlation vector matrix between entity references and the relationship matrix corresponding to the central entity reference of the answer library specifically includes the following process:

[0041] ;

[0042] in, is the module similarity, is the number of vectors in the correlation vector matrix between entity references, Represents the correlation vector matrix and the relationship matrix adjacency matrix, , represents the node number of the adjacency matrix, Represents the adjacency matrix The sum of all edges in , express The degree, express The degree, Representation node The community that exists, Representation node The community that exists, is the conditional restriction coefficient. If the correlation vector matrix between entity references and the number of entity reference characters in the relationship matrix corresponding to the central entity reference in the answer library are different, then Is 0, if the correlation vector matrix between entity references and the relationship matrix entity reference number corresponding to the central entity reference of the answer library, then is 1.

[0043] Furthermore, the face photo is an RGB image.

[0044] Compared with the existing solutions, the present invention achieves the following beneficial effects:

[0045] The present invention activates a face photo acquisition system based on question voice and collects the face photo of the current user, extracts target features representing the face from the face photo, calculates similarity between the target features and high-dimensional features stored in a question-and-answer client, and judges the legitimacy of the user based on the similarity; substitutes the question voice of the legitimate user into a voice-to-text recognition model, and outputs the voice content text; calculates the module similarity between the correlation vector matrix between entity references and the relationship matrix corresponding to the central entity reference of the answer library, and outputs the preset answer text corresponding to the central entity reference with the largest module similarity, thereby improving the personalization, accuracy, and intelligence level of question-and-answer, improving the efficiency of information acquisition, and enhancing the scalability and flexibility of the system.

[0046] By leveraging natural language processing and deep learning technologies, the system can more accurately understand the user's question intent and reduce misunderstandings and ambiguity. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0048] Figure 1 This is a system block diagram of a personalized intelligent question-answering system based on artificial intelligence according to an embodiment of the present invention;

[0049] Figure 2 This is a workflow diagram of the first artificial intelligence-based personalized intelligent question-answering system according to an embodiment of the present invention;

[0050] Figure 3 This is a workflow diagram of the second artificial intelligence-based personalized intelligent question-answering system in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] In addition, the described features, structures or characteristics can be combined in any suitable manner in one or more example embodiments. In the following description, many specific details are provided to provide a full understanding of the example embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure can be practiced while omitting one or more of the specific details, or other methods, components, steps, etc. can be adopted. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0053] This embodiment provides a personalized intelligent question-answering system based on artificial intelligence. Figure 1 This is a system block diagram of a personalized intelligent question-answering system based on artificial intelligence according to an embodiment of the present invention. Figure 1 As shown, the system includes:

[0054] The legitimacy judgment module is used to obtain the user's voice question, activate the face photo collection system based on the question voice, collect the current user's face photo, extract the target features representing the face from the face photo, calculate the similarity between the target features and the high-dimensional features stored in the question-answering client, and judge the legitimacy of the user based on the similarity;

[0055] The speech-to-text recognition module is used to substitute the legitimate user's question speech into the speech-to-text recognition model and output the speech content text;

[0056] The entity reference extraction module is used to extract entity references from the speech content text, determine the correlation between entity references in the speech content text based on association rule mining, perform PCA analysis on the associated entity references, and obtain the correlation vector matrix between the associated entity references;

[0057] The answer text output module is used to calculate the module similarity between the correlation vector matrix between entity references and the relationship matrix corresponding to the central entity reference in the answer library, and output the preset answer text corresponding to the central entity reference with the largest module similarity.

[0058] In summary, the present invention activates the face photo collection system based on the question voice and collects the face photo of the current user, extracts the target features representing the face from the face photo, calculates the similarity between the target features and the high-dimensional features stored in the question and answer client, and judges the legitimacy of the user based on the similarity; substitutes the question voice of the legitimate user into the voice-to-text recognition model, and outputs the voice content text; calculates the module similarity between the correlation vector matrix between entity references and the relationship matrix corresponding to the central entity reference of the answer library, and outputs the preset answer text corresponding to the central entity reference with the largest module similarity, thereby improving the personalization, accuracy, and intelligence level of question and answer, improving the efficiency of information acquisition, and enhancing the scalability and flexibility of the system.

[0059] It is worth noting that the facial photo acquisition system usually consists of front-end acquisition equipment, data transmission module, data processing center and user interaction interface, and the facial photos it collects are RGB images.

[0060] In some embodiments, calculating the similarity between the target feature and the high-dimensional feature stored in the question-answering client specifically includes the following process:

[0061] Calculate the basic similarity of the feature vector values ​​between the target feature and the high-dimensional feature:

[0062] ;

[0063] in, is the basic similarity, represents the i-th eigenvector value of the target feature representing the face, It represents the i-th feature vector value of the high-dimensional feature vector stored in the Q&A client, n is the dimension of the feature vector, and the high-dimensional features stored in the Q&A client are pre-extracted user facial features.

[0064] Further, Figure 2 This is a workflow diagram of the first personalized intelligent question-answering system based on artificial intelligence according to an embodiment of the present invention. Figure 2 As shown in FIG, judging the legitimacy of a user based on similarity specifically includes the following process:

[0065] Step S201: Obtain the question-answering time of the user's last question and the question-answering time of the current question and the answering time, calculate the time interval between the question-answering time of the last question and the answering time of the current question and the answering time, and record the time interval as the target time interval; sequentially calculate the historical similarity between all historical time intervals of the user and the query history time interval;

[0066] Step S202: Number the historical time intervals and the target time interval to obtain number values, construct a rectangular coordinate system with the number values ​​as the X-axis and the historical similarities and similarities as the Y-axis, connect adjacent similarities in the rectangular coordinate system to generate a similarity curve, draw perpendicular lines from both ends of the similarity curve to the X-axis to obtain two start and end line segments, and form a closed figure with the similarity curve, the two start and end line segments, and the X-axis, and calculate the total area of ​​the closed figure;

[0067] Step S203: Determine whether the total area exceeds a preset threshold. If so, determine that the user is a legitimate user. If not, determine that the user is an illegal user and deny access to the question-answering system.

[0068] In some embodiments, substituting the legitimate user's question speech into the speech-to-text recognition model and outputting the speech content text specifically includes the following process:

[0069] For problem speech data , is a set of problem speech data samples;

[0070] The speech content recognition model consists of two parts: the speech content encoding network f and the speech content context network. The speech content encoder network f consists of a five-layer convolutional network. The speech content encoder outputs a low-frequency feature. , the hidden space after the encoder network mapping is represented by Z, where :

[0071] ;

[0072] Among them, m is the convolution receptive field;

[0073] The speech content context network is a convolutional neural network with the same structure as the speech content encoding network f. It maps the hidden space into the context information space C. The output of the context network is ,in, :

[0074] ;

[0075] Among them, v is the convolution receptive field;

[0076] Loss function based on speech content recognition model To train the speech content recognition model, the loss function It is expressed as follows:

[0077] ;

[0078] in, is the sequence length of the problem speech data sample set X, is the loss of noise contrast estimation, is the true probability of the speech data sample to be evaluated, is the learning parameter;

[0079] Output the speech content text.

[0080] In some embodiments, entity reference extraction from speech content text based on the BERT (Bidirectional Encoder Representations from Transformers) model specifically includes the following process:

[0081] Text preprocessing:

[0082] The converted text is preprocessed, including word segmentation, stop word removal, punctuation processing, etc. This step is to enable the BERT model to better understand and process the text.

[0083] BERT model loading and fine-tuning:

[0084] Load the pre-trained BERT model and fine-tune it based on the entity reference extraction task. During the fine-tuning process, you can use a dataset with entity reference annotations for training, allowing the BERT model to learn specific patterns for entity reference extraction.

[0085] Entity reference extraction:

[0086] Use the fine-tuned BERT model to extract entity references from text. The BERT model outputs vector representations for each word or phrase, as well as the semantic relationships between them. By comparing these vector representations with the semantic relationships, we can identify entity references in the text.

[0087] In some embodiments, Figure 3 This is a workflow diagram of the second personalized intelligent question-answering system based on artificial intelligence according to an embodiment of the present invention. Figure 3 As shown, PCA analysis is performed on the associated entity references to obtain the correlation vector matrix between the associated entity references, which specifically includes the following process:

[0088] Step S301: forming a matrix M with associated entity references, and calculating the covariance matrix S of the samples of the matrix M;

[0089] Step S302: Calculate the eigenvectors and eigenvalues ​​of the covariance matrix S; where the eigenvector is 、 …、 , eigenvalues ​​t = 1,2,…, ;

[0090] Step S303: Projecting data into the feature vector space;

[0091] Using the formula:

[0092] ;

[0093] in, The value is the feature value corresponding to the associated entity reference in the original sample;

[0094] Step S304: Select R vectors and form a correlation vector matrix.

[0095] In some embodiments, calculating the module similarity requirement between the correlation vector matrix between entity references and the relationship matrix corresponding to the central entity reference in the answer library specifically includes the following process:

[0096] ;

[0097] in, is the module similarity, is the number of vectors in the correlation vector matrix between entity references, Represents the correlation vector matrix and the relationship matrix adjacency matrix, , represents the node number of the adjacency matrix, Represents the adjacency matrix The sum of all edges in , express The degree, express The degree, Representation node The community that exists, Representation node The community that exists, is the conditional restriction coefficient. If the correlation vector matrix between entity references and the number of entity reference characters in the relationship matrix corresponding to the central entity reference in the answer library are different, then Is 0, if the correlation vector matrix between entity references and the relationship matrix entity reference number corresponding to the central entity reference of the answer library, then is 1.

[0098] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in the embodiments of this application are fully or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0099] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0100] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is only for some logical functions. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0102] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0103] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A personalized intelligent question-answering system based on artificial intelligence, characterized by: The system includes: The legitimacy judgment module is used to obtain the user's voice question, activate the face photo collection system based on the question voice, collect the current user's face photo, extract the target features representing the face from the face photo, calculate the similarity between the target features and the high-dimensional features stored in the question-answering client, and judge the legitimacy of the user based on the similarity; The process of judging the legitimacy of a user based on similarity specifically includes the following: Obtain the time of the user's last question and answer and the time of the current question and answer, calculate the time interval between the last question and answer and the current question and answer, and record this time interval as the target time interval; sequentially calculate the historical similarity between all historical time intervals of the user and the query history time interval; Number the historical time intervals and the target time interval to obtain numbered values. A rectangular coordinate system is constructed with the numbered values ​​as the X-axis and the historical similarity and similarity as the Y-axis. Adjacent similarities in the rectangular coordinate system are connected to generate a similarity curve. A perpendicular line is drawn from both ends of the similarity curve to the X-axis to obtain two start and end line segments. A closed figure is formed by the similarity curve, the two start and end line segments, and the X-axis. The total area of ​​the closed figure is calculated. Determine whether the total area exceeds a preset threshold. If so, the user is considered a legitimate user. If not, the user is considered an illegal user and is denied access to the question-answering system. The speech-to-text recognition module is used to substitute the legitimate user's question speech into the speech-to-text recognition model and output the speech content text: For problem speech data , is a set of problem speech data samples; The speech content recognition model consists of two parts: the speech content encoding network f and the speech content context network. The speech content encoder network f consists of a five-layer convolutional network. The speech content encoder outputs a low-frequency feature. , the hidden space after the encoder network mapping is represented by Z, where : ； Among them, m is the convolution receptive field; The speech content context network is a convolutional neural network with the same structure as the speech content encoding network f. It maps the hidden space into the context information space C. The output of the context network is ,in, : ； Among them, v is the convolution receptive field; Loss function based on speech content recognition model To train the speech content recognition model, the loss function It is expressed as follows: ； in, is the sequence length of the problem speech data sample set X, is the loss of noise contrast estimation, is the true probability of the speech data sample to be evaluated, is the learning parameter; Output voice content text; The entity reference extraction module is used to extract entity references from the speech content text, determine the correlation between entity references in the speech content text based on association rule mining, perform PCA analysis on the associated entity references, and obtain the correlation vector matrix between the associated entity references; The answer text output module is used to calculate the module similarity between the correlation vector matrix between entity references and the relationship matrix corresponding to the central entity reference in the answer library, and output the preset answer text corresponding to the central entity reference with the largest module similarity; The module similarity requirement between the correlation vector matrix between entity references and the relationship matrix corresponding to the central entity reference in the answer library specifically includes the following process: ； in, is the module similarity, is the number of vectors in the correlation vector matrix between entity references, Represents the correlation vector matrix and the relationship matrix adjacency matrix, , represents the node number of the adjacency matrix, Represents the adjacency matrix The sum of all edges in , express The degree, express The degree, Representation node The community that exists, Representation node The community that exists, is the conditional restriction coefficient. If the correlation vector matrix between entity references and the number of entity reference characters in the relationship matrix corresponding to the central entity reference in the answer library are different, then Is 0, if the correlation vector matrix between entity references and the relationship matrix entity reference number corresponding to the central entity reference of the answer library, then is 1.

2. The personalized intelligent question-answering system based on artificial intelligence according to claim 1, characterized in that: The specific process of calculating the similarity between the target feature and the high-dimensional feature stored in the question-answering client is as follows: Calculate the basic similarity of the feature vector values ​​between the target feature and the high-dimensional feature: ； in, is the basic similarity, represents the i-th eigenvector value of the target feature representing the face, It represents the i-th feature vector value of the high-dimensional feature vector stored in the Q&A client, n is the dimension of the feature vector, and the high-dimensional features stored in the Q&A client are pre-extracted user facial features.

3. The personalized intelligent question-answering system based on artificial intelligence according to claim 1, characterized in that: Entity reference extraction from speech content text specifically includes: entity reference extraction from speech content text based on the BERT model.

4. The personalized intelligent question-answering system based on artificial intelligence according to claim 1, characterized in that: The PCA analysis of the associated entity references to obtain the correlation vector matrix between the associated entity references specifically includes the following process: (1) Form the matrix M of the associated entity references and calculate the covariance matrix S of the samples of the matrix M; (2) Calculate the eigenvectors of the covariance matrix S 、 …、 and eigenvalues ​​t = 1,2,…, ; (3) Project the data into the feature vector space, using the formula: ； in, The value is the feature value corresponding to the associated entity reference in the original sample; (4) Select R vectors and form a correlation vector matrix.

5. The personalized intelligent question-answering system based on artificial intelligence according to claim 1 is characterized in that: Face photos are RGB images.

Images