Data search method and system based on artificial intelligence

Abstract

The invention discloses a data search method and system based on artificial intelligence, and relates to the technical field of information retrieval, and the method comprises the steps: receiving a user query, carrying out query analysis, and calculating the similarity between the query analysis and a data point feature vector; constructing a distance matrix, constructing a simple form, and calculating topological characteristics of the simple form through a boundary operator; and converting the topological features into topological feature vectors, fusing the topological feature vectors with the data point feature vectors, constructing a deep neural network DNN, and generating a search result. The method comprises the following steps: receiving a user inquiry, carrying out inquiry analysis, and calculating the similarity with a data point feature vector; constructing a distance matrix, constructing a simple form, and calculating topological characteristics of the simple form through a boundary operator; the topological features are converted into topological feature vectors to be fused with the data point feature vectors, a deep neural network DNN is constructed, and a search result is generated; the discrimination capability of similarity difference is obviously improved, the efficiency of data processing and query response is enhanced, and the accuracy and correlation of search results are improved.

Description

Technical Field

[0001] The present invention relates to the field of information retrieval technology, and in particular to a data search method and system based on artificial intelligence. Background Art

[0002] With the rapid development of information technology, the scale of data generation and storage has increased dramatically, especially the popularization of multimodal data, which has brought unprecedented challenges and opportunities to data search and information retrieval technology. Traditional data search methods are mostly focused on structured data and use keyword-based retrieval strategies, which cannot effectively process unstructured or semi-structured data forms. With the continuous advancement of artificial intelligence technology, especially the rise of deep learning and natural language processing technology, researchers have gradually applied these advanced technologies to the processing and search of multimodal data.

[0003] Existing data search methods still face many challenges. When fusing features, most existing methods rely only on simple weighted averaging or concatenation, and fail to effectively utilize the deep relationship between the features of each modality, resulting in insufficient feature representation capabilities, which in turn affects the accuracy and recall of the search. For models based on deep learning, insufficient historical data during model training and the lack of optimization strategies further limit the effectiveness of the search algorithm. Summary of the invention

[0004] In view of the problems existing in the above-mentioned existing artificial intelligence-based data search methods, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that most of the existing methods only rely on simple weighted averaging or splicing when fusing features, and fail to effectively utilize the deep relationship between the features of each modality, resulting in insufficient feature representation capabilities, which in turn affects the accuracy and recall of the search. For models based on deep learning, insufficient historical data in the model training process and the lack of optimization strategies further limit the effectiveness of the search algorithm.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: an artificial intelligence-based data search method and system, which includes collecting multimodal data for preprocessing, extracting feature vectors of preprocessed data points, receiving user inquiries for query analysis and calculating the similarity between the data point feature vectors; constructing a distance matrix, constructing a simple shape according to the distance matrix, and calculating the topological features of the simple shape through a boundary operator; converting the topological features into topological feature vectors and fusing them with data point feature vectors, constructing a deep neural network DNN to calculate the similarity between the fused feature vector and the query feature vector, and generating search results; and constructing a visualization interface to display the optimized search results.

[0007] As a preferred solution of the artificial intelligence-based data search method of the present invention, the collecting of multimodal data for preprocessing and extracting feature vectors of preprocessed data points includes: Collect text data, voice data, and video data through the API interface and pre-process them separately; The preprocessing includes using regular expressions to delete HTML tags and special symbols from text data, using a bandpass filter to perform denoising on voice data, and using a Gaussian filter to remove noise from video data; Use the BERT model to convert text data into text feature vectors; The speech data is framed to obtain a speech signal, the power spectrum of the speech signal is calculated using Fourier transform, the MFCC features of the speech data are extracted using a Mel filter, and the MFCC features are concatenated to obtain a speech feature vector; Use VGG to extract the spatial features of video data, and combine LSTM to process the spatial features to obtain the video feature vector; The feature vector of the data point Perform standardization to obtain the kth data point feature vector after standardization ; The data point feature vector includes a text feature vector, a voice feature vector and a video feature vector.

[0008] As a preferred solution of the artificial intelligence-based data search method of the present invention, the receiving of user inquiries, performing query parsing and calculating the similarity with the feature vector of the data point includes: Receive user query information, load the pre-trained BERT model to generate the query feature vector U ’ Normalize the query feature vector; Use cosine similarity to calculate the normalized query feature vector U and the normalized feature vector of the kth data point Similarity between .

[0009] As a preferred solution of the artificial intelligence-based data search method of the present invention, wherein: constructing a distance matrix, constructing a simple shape according to the distance matrix, and calculating the topological features of the simple shape by a boundary operator include: Use Euclidean distance to calculate the distance between feature vectors of data points ; The distances between all feature vectors form a distance matrix D; Extract the non-zero distance values ​​in the distance matrix D and sort them from small to large to generate a sorted distance sequence. Use the percentile method to extract the minimum distance of the distance sequence to set the initial distance threshold. Set the incremental step size based on the difference between the distance sequences. Use the stepwise increment method to gradually increase the distance threshold according to the distance sequence. Use the distance threshold to gradually build a simple shape. ; Calculate the topological features of the a-th dimension of the simple shape through the boundary operator .

[0010] As a preferred solution of the artificial intelligence-based data search method of the present invention, the step of converting the topological features into a topological feature vector and fusing it with the data point feature vector comprises: Use persistent entropy to convert topological features into topological feature vector E i ’ And perform standardization, use weighted pooling to normalize the standardized topological feature vector E i Combined into a comprehensive topological feature vector E; The integrated data point feature vector F of the feature vector of the normalized data is calculated using weighted average; The comprehensive topological feature vector E and the kth data point feature vector after normalization are calculated by nonlinear transformation The fused feature vector between , the formula is: ,in is the hyperbolic tangent function, is the minimum value, is the dot product of the kth data point feature vector and the comprehensive topological feature vector after standardization, and are the L2 norms of the kth data point feature vector and the comprehensive topological feature vector after standardization, respectively.

[0011] As a preferred solution of the artificial intelligence-based data search method of the present invention, wherein: the construction of a deep neural network DNN to calculate the similarity between the fusion feature vector and the query feature vector and generate search results includes: Collect historical data, preprocess and extract historical feature vectors to generate training sets; Build a deep neural network DNN model, including input layer, hidden layer and output layer; Set the format of the input layer to be a fused feature vector and the normalized query feature vector U; Use the training set to train the deep neural network DNN model; Use the adaptive adjustment mechanism to improve the existing loss function and construct an improved loss function to optimize the model parameters. The formula is: ,in is the improved loss function, is the dot product of the normalized query feature vector and the fused feature vector, and are the L2 norms of the normalized query feature vector and fusion feature vector, w is the number of samples, is the output of the deep neural network DNN model, is the true label, is the Euclidean distance between the normalized query feature vector and the fused feature vector; The fused feature vector and the query feature vector are introduced into the trained deep neural network DNN model to obtain the fused feature vector The similarity between the normalized query feature vector U ; The fused feature vector The similarity between the normalized query feature vector U The query feature vector U after normalization and the kth data point feature vector after normalization Similarity between Compare and compare the similarity Greater than similarity The feature vector of the data point is retained, and the corresponding multimodal data is extracted to generate the final search result.

[0012] As a preferred solution of the artificial intelligence-based data search method of the present invention, the step of constructing a visual interface to display optimized search results includes: Use React.js to build a visual interface to display user query information and final search results, and use D3.js to dynamically display multimodal data; Users who have passed real-name verification are allowed to view the information.

[0013] Another object of the present invention is to provide an artificial intelligence-based data search system, which includes.

[0014] A computer device comprises: a memory and a processor; the memory stores a computer program, and the processor implements the steps of the above-mentioned artificial intelligence-based data search method when executing the computer program.

[0015] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned artificial intelligence-based data search method.

[0016] The beneficial effects of the present invention are as follows: the present invention collects multimodal data for preprocessing, extracts feature vectors of preprocessed data points, receives user inquiries for query analysis and calculates the similarity with the feature vectors of data points; constructs a distance matrix, constructs a simple shape according to the distance matrix, and calculates the topological features of the simple shape through a boundary operator; converts the topological features into topological feature vectors and fuses them with data point feature vectors, constructs a deep neural network DNN to calculate the similarity between the fused feature vector and the query feature vector, and generates search results; significantly improves the ability to distinguish similarity differences, enhances the efficiency of data processing and query response, and improves the accuracy and relevance of search results. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 Schematic diagram of the process of data search method based on artificial intelligence.

[0019] Figure 2 This is a schematic diagram of the structure of the data search system based on artificial intelligence. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, reference Figure 1 , which is the first embodiment of the present invention, and which provides a data search method based on artificial intelligence. The data search method based on artificial intelligence includes: S1, collect multimodal data for preprocessing, extract feature vectors of preprocessed data points, receive user inquiries, perform query parsing, and calculate the similarity between the data point feature vectors; Specifically, multimodal data is collected for preprocessing, and feature vectors of preprocessed data points are extracted, including: Collect text data, voice data, and video data through the API interface and pre-process them separately; The preprocessing includes using regular expressions to delete HTML tags and special symbols from text data, using a bandpass filter to perform denoising on voice data, and using a Gaussian filter to remove noise from video data; Use the BERT model to convert text data into text feature vectors; The speech data is framed to obtain a speech signal, the power spectrum of the speech signal is calculated using Fourier transform, the MFCC features of the speech data are extracted using a Mel filter, and the MFCC features are concatenated to obtain a speech feature vector; Use VGG to extract the spatial features of video data, and combine LSTM to process the spatial features to obtain the video feature vector; The feature vector of the data point Perform standardization to obtain the kth data point feature vector after standardization ; The data point feature vector includes a text feature vector, a voice feature vector and a video feature vector.

[0024] By collecting multimodal data from different sources through the API interface, the system can achieve unified management and processing of data. This integration capability ensures the comprehensiveness and accuracy of data analysis, and provides a solid data foundation for subsequent searches and queries. Multimodal data that has been effectively denoised and formatted can more accurately reflect real information and reduce errors in subsequent model processing. By using advanced technologies such as the BERT model, Mel filter and LSTM, the present invention can extract feature vectors with rich semantic and contextual information. These high-quality feature vectors will significantly improve the accuracy and relevance of the search engine during query, thereby improving the user experience. The standardized data point feature vectors can effectively shorten the calculation time of subsequent queries and analyses, and provide a feasible solution for real-time data analysis and rapid response application scenarios.

[0025] Further, receiving the user query, performing query parsing and calculating the similarity with the feature vector of the data point includes: Receive user query information, load the pre-trained BERT model to generate the query feature vector U ’ The query feature vector is normalized to obtain a normalized query feature vector U; Use cosine similarity to calculate the normalized query feature vector U and the normalized data point feature vector Similarity between .

[0026] By loading the pre-trained BERT model, the system can quickly understand the user's query intent. The efficient understanding ability can provide users with more accurate and relevant search results, improving the user experience. The generation of feature vectors is combined with standardization processing to ensure that query information and data points are compared on the same scale. The standardization process significantly improves the accuracy of cosine similarity calculation, allowing the system to more effectively identify relevant data points. Using cosine similarity to calculate the similarity between query feature vectors and data point feature vectors can quickly locate the data most relevant to the user's query. Accurate matching capabilities are particularly important for processing complex queries, especially in multimodal data environments.

[0027] S2, construct a distance matrix, construct a simple shape according to the distance matrix, and calculate the topological features of the simple shape through the boundary operator; Specifically, construct a distance matrix, construct a simple shape based on the distance matrix, use persistent homology to analyze the topological features of the simple shape, and plot the birth and death time of each topological feature into a persistent graph, including: Use Euclidean distance to calculate the distance between feature vectors of data points The distance matrix D is formed as follows: , Where n is the total number of eigenvectors; Extract the non-zero distance values ​​in the distance matrix D and sort them from small to large to generate a sorted distance sequence. Use the percentile method to extract the minimum distance of the distance sequence to set the initial distance threshold. Set the incremental step size based on the distance difference between the first and second distance sequence. Use the stepwise increment method to gradually increase the distance threshold according to the distance sequence. Use the distance threshold to gradually build a simple shape. The formula is: ,in To gradually increase the distance threshold The a-th dimension simple form constructed below, is the feature vector of a+1 data points involved in constructing the a-th dimension simple shape; Starting from the initial distance threshold, if the distance between each pair of eigenvectors is less than or equal to the gradually increasing distance threshold, the eigenvectors are connected to form a one-dimensional simple shape, and the line segments form a topological structure; If the distance between each pair of three eigenvectors is less than or equal to the gradually increasing distance threshold, the eigenvectors are connected to form a two-dimensional simple shape, a triangle; If the distance between each pair of four eigenvectors is less than or equal to the gradually increasing distance threshold, the eigenvectors are connected to form a three-dimensional simple shape, a tetrahedron; … If the distance between each pair of n eigenvectors is less than or equal to the gradually increasing distance threshold, the eigenvectors are connected to form an n-1 dimensional simple shape; Calculate the topological features of the a-th dimension of the simple shape through the boundary operator , the formula is: , ,in is the boundary operator of the a-th dimension simplex, is the sign term, which is used to control the sign in the summation. is the kernel of the a-th dimension boundary operator, capturing the topological structure, It is the boundary image of the a+1th dimensional simple shape, removing the non-independent topological structure.

[0028] By using Euclidean distance to calculate the distance between the feature vectors of data points, the accuracy of the distance matrix is ​​ensured. Accurate distance measurement is the basis for subsequent analysis, making the similarity assessment of data more reliable. The simple shape is constructed by the step-by-step incremental method, which can effectively reflect the topological characteristics of the data set. The construction of the simple shape makes high-dimensional data visualized during the analysis process, which promotes the understanding of the data structure. The initial distance threshold and incremental step size are set so that the method can flexibly adapt to the characteristics of different data sets. This adaptability ensures the effectiveness of the algorithm in various application scenarios, and it can perform well in both dense and sparse data environments. In a multimodal data environment, the construction of the distance matrix can handle complex data relationships, which enables the system to When processing multimodal data such as text, voice and video, it still maintains high efficiency and accuracy. By extracting topological features in the distance matrix, it can reveal the potential structure and pattern in the data. This process is crucial to data mining and knowledge discovery, helping researchers and companies identify key data points and their relationships. The weighted merging and sorting processing of the distance matrix significantly reduces the complexity of calculations. By optimizing the algorithm design, the present invention improves the computing efficiency while ensuring accuracy, and is suitable for real-time data processing needs. Through a deep understanding of the relationship between data points and the extraction of topological features, the present invention provides users with powerful decision-making support. Whether in recommendation systems, classification models or predictive analysis, accurate topological information can effectively improve the scientific nature of decision-making.

[0029] S3, converting the topological features into topological feature vectors and fusing them with data point feature vectors, constructing a deep neural network DNN to calculate the similarity between the fused feature vectors and the query feature vectors, and generating search results; Specifically, converting the topological features into a topological feature vector and fusing it with the data point feature vector includes: Use persistent entropy to convert topological features into topological feature vector E i ’ And perform standardization, use weighted pooling to normalize the standardized topological feature vector E i Combined into a comprehensive topological feature vector E; Based on the existing technology, linear weighted fusion is used to calculate the comprehensive topological feature vector E and the kth data point feature vector after standardization The fused feature vector between , the formula is: ,in and are weighting coefficients respectively; The comprehensive topological feature vector E and the kth data point feature vector after normalization are calculated by nonlinear transformation The fusion feature vector , the formula is: ,in is the hyperbolic tangent function, which is used to nonlinearly compress the similarity between the topological eigenvector and the data point eigenvector. is a minimum value, which is used to avoid division by zero errors or very small values ​​being ignored, and to ensure numerical stability during the calculation process. is the dot product of the kth data point feature vector and the comprehensive topological feature vector after standardization, and are the L2 norms of the kth data point feature vector and the comprehensive topological feature vector after normalization; The hyperbolic tangent function tanh nonlinear activation function can compress the eigenvalues ​​and limit the output range to −1 to 1. It can capture the nonlinear relationship between features. Linear methods such as simple weighting or addition may lead to oversimplification of the relationship between features and fail to fully express the structure of complex data. The nonlinear activation function can better retain the subtle differences between features, especially in scenarios with large data volumes. By introducing nonlinear transformations, the model's ability to capture the relationship between complex features is enhanced, which is particularly suitable for high-dimensional and complex data sets. The dot product can be used to calculate the similarity of two vectors in the same direction to ensure that the comprehensive features are well preserved. The maximum correlation is captured in the mutual comparison between the eigenvectors. The normalization operation after the dot product further eliminates the influence of the vector length, so that the fusion result only reflects the direction and similarity of the two eigenvectors. The normalization operation ensures that the fusion process will not be unbalanced due to the vector modulus being too large or too small. The summation operation is often used to calculate the sum of multiple eigenvectors, but this may increase the computational complexity, especially in the multi-dimensional feature space. By first synthesizing the eigenvectors and then fusing them, the complex element-by-element summation calculation is avoided and the calculation process is simplified. This method not only improves efficiency, but also ensures the balance of feature fusion.

[0030] Persistent entropy can quantify the complexity and persistence of topological features, and is particularly suitable for extracting topological structures from high-dimensional data. The standardized topological feature vector can eliminate the scale differences between features, so that the topological features have a consistent expression in the subsequent fusion process. Standardization can also prevent features of a specific dimension from excessively affecting the final result during fusion, thereby improving the reliability and accuracy of feature fusion. Nonlinear transformation using the hyperbolic tangent function can effectively compress the similarity results between feature vectors, avoid the influence of extreme values, increase the nonlinear expression ability of the model, and enable complex data relationships to be more fully captured. The use of nonlinear transformation can also ensure the correlation between features. The similarity is mapped to the output space more smoothly, which enhances the generalization ability of the model. The construction of fused feature vectors improves the generalization ability of the model for unknown data. Through the effective fusion of different types of features, the model can still maintain a high prediction accuracy when facing new data. The fused feature vectors can provide more comprehensive information for the decision-making system, helping analysts to obtain more accurate reference basis when making data-driven decisions. The automated feature fusion method significantly reduces the complexity of manual feature engineering, making data analysis more efficient. This feature is especially important for start-ups or teams with limited resources, and can help them obtain results more quickly in data analysis.

[0031] Furthermore, a deep neural network DNN is constructed to calculate the similarity between the fusion feature vector and the query feature vector, and the search results generated include: Collect historical data, preprocess and extract historical feature vectors to generate training sets; Build a deep neural network DNN model, including input layer, hidden layer and output layer; Set the format of the input layer to be a fused feature vector and the normalized query feature vector U; Use the training set to train the deep neural network DNN model and use the existing loss function , the formula is: , where w is the number of samples, is the output of the deep neural network DNN model, is the true label; Use the adaptive adjustment mechanism to improve the existing loss function and construct an improved loss function to optimize the model parameters. The formula is: ,in is the improved loss function, is the dot product of the normalized query feature vector and the fused feature vector, and are the L2 norms of the normalized query feature vector and the fused feature vector, indicating the modulus of the vector, is the Euclidean distance between the query feature vector and the fusion feature vector, w is the number of samples, is the output of the deep neural network DNN model, is the true label, is the Euclidean distance between the normalized query feature vector and the fused feature vector; In the prior art, cosine similarity is usually used directly to measure the similarity between two vectors. The change of the original cosine similarity is relatively smooth when it is close to 1. When the similarity of the two vectors is close, the difference is not obvious. The model is not sensitive enough to such small changes and may not be able to effectively distinguish vectors with similarities that are close but not exactly the same. The use of square difference enhances the model's sensitivity to similarity differences. Small differences will be amplified, ensuring that the model more actively optimizes the similarity of feature vectors and significantly improves the ability to distinguish similarity differences. Cosine similarity only considers the directional similarity of the two vectors and ignores the absolute distance between them. The two vectors may be very similar in direction but very different in magnitude. The introduction of the Euclidean distance model reduces the similarity score when processing vectors with similar directions but far distances. The model can dynamically adjust the loss value to ensure The model considers the absolute distance of the vectors and does not rely solely on directional similarity. MSE is the optimal option when processing continuous value predictions and cannot be replaced by other simple error metrics such as MAE. Normalization increases the denominator when extremely large errors occur, thereby reducing the contribution of these points to the total loss. MAE maintains the MSE amplification effect on large errors while also reducing the impact of outliers. If the cosine similarity and MSE losses are simply added together, the model cannot adaptively adjust the influence of each part according to different scenarios. Through the design of the fractional structure and the normalization mechanism of each loss term, the model can be ensured to adaptively adjust the weights of each loss term, avoiding the limitations of traditional linear combinations. The fractional structure realizes adaptive adjustment, and each loss term is dynamically adjusted through its own normalization mechanism, ensuring that the loss optimization of the model in different scenarios is more balanced and reasonable.

[0032] The fused feature vector and the query feature vector are introduced into the trained deep neural network DNN model to obtain the fused feature vector The similarity between the normalized query feature vector U ; The fused feature vector The similarity between the normalized query feature vector U The query feature vector U after normalization and the kth data point feature vector after normalization Similarity between Compare and compare the similarity Greater than similarity The feature vector of the data point is retained, and the corresponding multimodal data is extracted to generate the final search result; The collecting of historical data and preprocessing and extracting of historical feature vectors include: using the same method as preprocessing and feature extraction of multimodal data to construct a historical distance matrix, constructing a historical simple shape based on the historical distance matrix, calculating the topological features of the historical simple shape through a boundary operator, converting the historical topological features into a historical topological feature vector and fusing it with a historical data point feature vector to obtain a historical fused feature vector.

[0033] By calculating the similarity between the fusion feature vector and the query feature vector, the present invention can effectively improve the accuracy and relevance of search results. Compared with traditional methods, the model based on deep learning can better capture the complex relationships in the data, thereby providing more personalized and accurate search results. By constructing a deep neural network and using an adaptive adjustment mechanism to optimize the loss function, the model can adapt to different data distributions, thereby performing more stably on new data. This feature enables the present invention to maintain a high accuracy when facing dynamic data. By collecting historical data and extracting feature vectors, the model can make full use of existing data resources, thereby enhancing the practicality and effectiveness of the model in practical applications. The use of data not only improves the learning effect of the model, but also reduces the cost of data collection and processing, allowing users to find the required information more quickly. The deep neural network model can improve efficiency through parallel computing when processing large-scale data. The processing of fused feature vectors can significantly reduce the consumption of computing resources, making it more efficient in practical applications. Search results based on real-time relevance scoring can provide a more intelligent user interaction experience. Users can get search results that better meet their needs, thereby improving satisfaction and usage stickiness. By exploring the application of deep neural networks in feature fusion and similarity, the present invention provides a new research direction for the field of machine learning and promotes the progress and development of related technologies.

[0034] S4. Build a visual interface to display the optimized search results; Specifically, building a visual interface to display optimized search results includes: using React.js to build a visual interface to display user query information and final search results, and using D3.js to dynamically display multimodal data; Users who have passed real-name verification are allowed to view the information.

[0035] The interface built with React.js provides an efficient user interaction experience. The componentized structure enables the interface to quickly respond to user operations, improving the overall convenience and fluency of use. This dynamic response capability enhances user participation and improves user satisfaction. With the powerful drawing ability of D3.js, data points can be displayed in a dynamic way, allowing users to intuitively understand search results in a variety of graphical forms (such as line charts, bar charts, scatter plots, etc.). This intuitive visualization method helps users quickly identify data trends and patterns, so as to make more effective decisions. The design of the visual interface can convert complex data sets into easy-to-understand information, helping users to quickly find key information in massive data. By displaying data relationships in a graphical way, users can more clearly identify correlations and improve the efficiency of data analysis. Through the real-name verification mechanism, it is ensured that only verified users can access and query optimized search results. This security measure effectively protects user data and prevents unauthorized access, thereby enhancing the credibility of the system and the trust of users. The present invention combines the visual display of the front end with the search optimization technology of the back end, so that users can view the optimized search results in real time. This instant feedback mechanism can help users understand the search process and improve users' trust in the system.

[0036] Example 2, reference Figure 2 , which is the second embodiment of the present invention, and which is different from the previous embodiment, provides a data search system based on artificial intelligence, including: The collection and query module is used to collect multimodal data for preprocessing, extract the feature vector of the preprocessed data point, receive user inquiries for query parsing and calculate the similarity between the data point feature vector; Construct an analysis module to construct a distance matrix, construct a simple shape based on the distance matrix, and calculate the topological features of the simple shape through the boundary operator; Fusion optimization module, used to convert topological features into topological feature vectors and fuse them with data point feature vectors, build a deep neural network DNN to calculate the similarity between the fusion feature vector and the query feature vector, and generate search results; The visualization module is used to build a visualization interface to display the optimized search results.

[0037] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0038] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0039] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0040] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit with a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit with a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

Claims

1. A data search method based on artificial intelligence, characterized in that: include: Collect multimodal data for preprocessing, extract feature vectors of preprocessed data points, receive user inquiries for query parsing and calculate similarity with feature vectors of data points; Construct a distance matrix, construct a simple shape based on the distance matrix, and calculate the topological features of the simple shape through the boundary operator; The topological features are converted into topological feature vectors and fused with data point feature vectors, and a deep neural network DNN is constructed to calculate the similarity between the fused feature vector and the query feature vector to generate search results; Build a visual interface to display optimized search results.

2. The artificial intelligence-based data search method according to claim 1, characterized in that: The collecting of multimodal data for preprocessing and extracting feature vectors of preprocessed data points includes: Collect text data, voice data, and video data through the API interface and pre-process them separately; The preprocessing includes using regular expressions to delete HTML tags and special symbols from text data, using a bandpass filter to perform denoising on voice data, and using a Gaussian filter to remove noise from video data; Use the BERT model to convert text data into text feature vectors; The speech data is framed to obtain a speech signal, the power spectrum of the speech signal is calculated using Fourier transform, the MFCC features of the speech data are extracted using a Mel filter, and the MFCC features are concatenated to obtain a speech feature vector; Use VGG to extract the spatial features of video data, and combine LSTM to process the spatial features to obtain the video feature vector; The feature vector of the data point Perform standardization to obtain the kth data point feature vector after standardization ; The data point feature vectors include text feature vectors, voice feature vectors and video feature vectors.

3. The artificial intelligence-based data search method according to claim 2, characterized in that: The receiving of user query, performing query parsing and calculating the similarity with the data point feature vector comprises: Receive user query information, load the pre-trained BERT model to generate a query feature vector from the query information The query feature vector is normalized to obtain a normalized query feature vector U; Use cosine similarity to calculate the normalized query feature vector U and the normalized feature vector of the kth data point Similarity between .

4. The artificial intelligence-based data search method according to claim 3, characterized in that: The step of constructing a distance matrix, constructing a simple shape according to the distance matrix, and calculating the topological features of the simple shape by using a boundary operator includes: Use Euclidean distance to calculate the distance between feature vectors of data points ; The distances between all feature vectors form a distance matrix D; Extract the non-zero distance values ​​in the distance matrix D and sort them from small to large to generate a sorted distance sequence. Use the percentile method to extract the minimum distance of the distance sequence to set the initial distance threshold. Set the incremental step size based on the difference between the distance sequences. Use the stepwise increment method to gradually increase the distance threshold according to the distance sequence. Use the distance threshold to gradually build a simple shape. ; Calculate the topological features of the a-th dimension of the simple shape through the boundary operator .

5. The artificial intelligence-based data search method according to claim 4, characterized in that: The converting of the topological features into a topological feature vector and fusing it with the data point feature vector comprises: Use persistent entropy to convert topological features into topological feature vector E i ’ And perform standardization, use weighted pooling to normalize the standardized topological feature vector E i Combined into a comprehensive topological feature vector E; The comprehensive topological feature vector E and the kth data point feature vector after normalization are calculated by nonlinear transformation The fusion feature vector between , the formula is: ,in is the hyperbolic tangent function, is the minimum value, is the dot product of the kth data point feature vector and the comprehensive topological feature vector after standardization, and are the L2 norms of the kth data point feature vector and the comprehensive topological feature vector after standardization, respectively.

6. The artificial intelligence-based data search method according to claim 5, characterized in that: The constructing of a deep neural network DNN to calculate the similarity between the fusion feature vector and the query feature vector and generating search results includes: Collect historical data, preprocess and extract historical feature vectors to generate training sets; Build a deep neural network DNN model, including input layer, hidden layer and output layer; Set the format of the input layer to be a fused feature vector and the normalized query feature vector U; Use the training set to train the deep neural network DNN model; Use the adaptive adjustment mechanism to improve the existing loss function and construct an improved loss function to optimize the model parameters. The formula is: ,in is the improved loss function, is the dot product of the normalized query feature vector and the fused feature vector, and are the L2 norms of the normalized query feature vector and fusion feature vector, w is the number of samples, is the output of the deep neural network DNN model, is the true label, is the Euclidean distance between the normalized query feature vector and the fused feature vector; the fused feature vector and the query feature vector are brought into the trained deep neural network DNN model to obtain the fused feature vector The similarity between the normalized query feature vector U ; The fused feature vector The similarity between the normalized query feature vector U The query feature vector U after normalization and the kth data point feature vector after normalization Similarity between Compare and compare the similarity Greater than similarity The feature vector of the data point is retained, and the corresponding multimodal data is extracted to generate the final search result.

7. The artificial intelligence-based data search method according to claim 6, characterized in that: The construction of a visual interface to display optimized search results includes: Use React.js to build a visual interface to display user query information and final search results, and use D3.js to dynamically display multimodal data; Users who have passed real-name verification are allowed to view the information.

8. An artificial intelligence-based data search system based on the artificial intelligence-based data search method according to any one of claims 1 to 7, characterized in that: include: The collection and query module is used to collect multimodal data for preprocessing, extract the feature vector of the preprocessed data point, receive user inquiries for query parsing and calculate the similarity between the data point feature vector; Construct an analysis module to construct a distance matrix, construct a simple shape based on the distance matrix, and calculate the topological features of the simple shape through the boundary operator; Fusion optimization module, used to convert topological features into topological feature vectors and fuse them with data point feature vectors, build a deep neural network DNN to calculate the similarity between the fusion feature vector and the query feature vector, and generate search results; The visualization module is used to build a visualization interface to display the optimized search results.

9. A computer device comprising: A memory and a processor; the memory stores a computer program, wherein the processor implements the steps of the artificial intelligence-based data search method described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence-based data search method according to any one of claims 1 to 7 are implemented.

Images