Information search method and device, computer equipment and storage medium

Abstract

The invention discloses an information search method and device, computer equipment and a storage medium, and belongs to the technical field of artificial intelligence. According to the method, through intelligent keyword combination and weight evaluation, the accuracy and relevance of search and query are improved, and compared with traditional keyword matching search, the method is based on the pre-trained keyword weight model, the importance of different keyword combinations can be automatically evaluated, invalid or low-relevance query is reduced, and the search efficiency is improved. Meanwhile, a deep learning model is adopted to optimize search query statements, so that the search query statements are more in line with user intentions, the quality of search results is enhanced, the accuracy of information retrieval is improved, and the cost of manually screening and adjusting search keywords is reduced. In a mass data environment, the information method provided by the invention can be applied to business management program systems of financial science and technology, medical treatment, health, old-age care and the like, and more intelligent and efficient information search experience is provided for users.

Description

Technical Field

[0001] This application belongs to the fields of artificial intelligence technology and fintech, and specifically relates to an information search method, device, computer device, and storage medium. Background Art

[0002] In the era of information explosion, the amount of data has grown exponentially, making it increasingly difficult for researchers to quickly and accurately obtain the required content from the vast amount of information. Traditional information search methods mainly rely on manual searching, screening, and sorting, which not only consume time and effort but also easily miss key information or be interfered by redundant data, affecting the comprehensiveness and accuracy of research. In addition, although existing search engines can provide certain support, due to algorithm limitations, they often have difficulty understanding the deep needs of users, resulting in insufficient relevance or excessive noise in search results.

[0003] Taking insurance information search as an example, the amount of information is huge and complex, including policy terms, product details, laws and regulations, and market dynamics. Researchers and consumers face many challenges when searching for relevant information. Existing insurance information search solutions mainly rely on keyword matching and have difficulty accurately understanding the specific needs of users, resulting in search results often containing a large amount of irrelevant information or missing key content.

[0004] Therefore, there is an urgent need for a more intelligent and efficient information search method that can combine technologies such as natural language processing and big data analysis to improve the accuracy and efficiency of information retrieval and provide more targeted knowledge support for researchers. Summary of the Invention

[0005] The purpose of the embodiments of this application is to propose an information search method, device, computer device, and storage medium to provide a more intelligent and efficient information search method that can combine technologies such as natural language processing and big data analysis to improve the accuracy and efficiency of information retrieval.

[0006] To solve the above technical problems, the embodiments of this application provide an information search method, which adopts the following technical solutions:

[0007] An information search method, comprising:

[0008] Receiving an information search instruction and obtaining a sequence of search keywords uploaded by a user, where the sequence of search keywords includes a plurality of search keywords;

[0009] Randomly combining the search keywords in the sequence of search keywords to obtain a plurality of keyword combinations;

[0010] Inputting the plurality of keyword combinations into a pre-trained keyword weight model in sequence to obtain keyword weights corresponding to the respective keyword combinations;

[0011] Sort the keyword combinations according to the keyword weights, and determine the target keyword combination from the sorting results;

[0012] Generate a matching search query statement according to the target keyword combination;

[0013] Based on the search query statement, use a pre-trained deep learning model to perform information search and output the information search results.

[0014] To solve the above technical problems, an embodiment of the present application further provides an information search device, which adopts the following technical solutions:

[0015] An information search device, comprising:

[0016] A keyword acquisition module, configured to receive an information search instruction and acquire a sequence of search keywords uploaded by a user, wherein the sequence of search keywords includes a plurality of search keywords;

[0017] A keyword combination module, configured to randomly combine the search keywords in the sequence of search keywords to obtain a plurality of keyword combinations;

[0018] A weight generation module, configured to sequentially input a plurality of keyword combinations into a pre-trained keyword weight model to obtain the keyword weights corresponding to each keyword combination;

[0019] A weight sorting module, configured to sort the keyword combinations according to the keyword weights, and determine the target keyword combination from the sorting results;

[0020] A statement generation module, configured to generate a matching search query statement according to the target keyword combination;

[0021] An information search module, configured to perform information search using a pre-trained deep learning model based on the search query statement and output the information search results.

[0022] To solve the above technical problems, an embodiment of the present application further provides a computer device, which adopts the following technical solutions:

[0023] A computer device, comprising a memory and a processor, wherein computer-readable instructions are stored in the memory, and when the processor executes the computer-readable instructions, the steps of the information search method described in any one of the above are implemented.

[0024] To solve the above technical problems, an embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solutions:

[0025] A computer-readable storage medium stores computer-readable instructions thereon, and when the computer-readable instructions are executed by a processor, the steps of the information search method described in any one of the above are implemented.

[0026] Compared with the prior art, the embodiments of the present application mainly have the following beneficial effects:

[0027] The present application discloses an information search method, device, computer device, and storage medium, belonging to the field of artificial intelligence technology. Through intelligent keyword combination and weight evaluation, the present application improves the accuracy and relevance of search queries. Compared with traditional keyword matching searches, this method is based on a pre-trained keyword weight model, which can automatically evaluate the importance of different keyword combinations, reduce invalid or low-relevance queries, and improve search efficiency. At the same time, a deep learning model is used to optimize the search query statement to make it more in line with the user's intention, enhance the quality of search results, not only improve the accuracy of information retrieval, but also reduce the cost of manual screening and adjustment of search keywords. In a massive data environment, the information method provided by the present application can be applied to business management program systems such as fintech, healthcare, and elderly care, providing users with a more intelligent and efficient information search experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the solutions in the present application, the following will briefly introduce the drawings required for the description of the embodiments of the present application. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0029] Figure 1 Shows an exemplary system architecture diagram to which the present application can be applied;

[0030] Figure 2 Shows a flowchart of an embodiment of the information search method according to the present application;

[0031] Figure 3 Shows a schematic structural diagram of an embodiment of the information search device according to the present application;

[0032] Figure 4 Shows a schematic structural diagram of an embodiment of the computer device according to the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above drawings are intended to cover non-exclusive inclusion. The terms "first", "second", etc. in the specification and claims of this application or the above drawings are used to distinguish different objects and not to describe a specific order.

[0034] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The phrase appearing in various places in the specification is not necessarily referring to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive of other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0035] To enable those skilled in the art of this technology to better understand the solutions of this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings.

[0036] As Figure 1 shown, the system architecture 100 may include a terminal device 101, a network 102, and a server 103. The terminal device 101 may be a laptop computer 1011, a tablet computer 1012, or a mobile phone 1013. The network 102 is a medium for providing a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0037] Users may use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications may be installed on the terminal device 101, such as a web browser application, a shopping application, a search application, an instant messaging tool, an email client, a social platform software, etc.

[0038] The terminal device 101 can be various electronic devices with a display screen and supporting web browsing. In addition to the laptop 1011, tablet computer 1012, or mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV) player, a laptop computer, a desktop computer, etc.

[0039] The server 103 can be a server that provides various services, such as a background server that supports the pages displayed on the terminal device 101.

[0040] It should be noted that the information search method provided by the embodiments of the present application is generally executed by the server / terminal device. Correspondingly, the information search device is generally set in the server / terminal device.

[0041] It should be understood that Figure 1 the numbers of the terminal devices, networks, and servers in

[0042] Continuing to refer to Figure 2 , a flowchart of an embodiment of the information search method according to the present application is shown. The information search method includes the following steps:

[0043] S201, receive an information search instruction, and obtain the search keyword sequence uploaded by the user, where the search keyword sequence includes several search keywords;

[0044] Specifically, the system receives the user's search requirement and extracts the input search keyword sequence. The user can input the search instruction in different ways, such as text input, voice recognition, or the system automatically generating a keyword list. To improve the search experience, the system can provide an auto-complete or keyword recommendation function to help the user optimize the search input. After receiving the search keyword sequence, the system will perform preprocessing, including removing stop words (such as "of" and "is"), word segmentation (such as using Jieba for Chinese), synonym expansion (such as classifying "auto insurance" and "motor vehicle insurance" as the same category), etc., to ensure the integrity and accuracy of the keywords. In addition, to improve the search flexibility, the system can automatically recommend possible search keywords based on the user's historical search behavior, industry popular keywords, or relevance analysis to form a more comprehensive search keyword sequence.

[0045] S202. Randomly combine the search keywords in the search keyword sequence to obtain a number of keyword combinations;

[0046] Specifically, after obtaining the user's search keyword sequence, the system will combine the keywords according to a certain strategy to generate multiple possible search query structures. Common combination methods include full permutation combination (listing all possibilities), fixed collocation (selecting high-frequency combinations based on industry knowledge), keyword clustering (combining based on semantic similarity), etc. The purpose of random combination is to expand the search coverage and ensure the diversity of queries, so as to increase the possibility of the search engine returning highly relevant results.

[0047] For example, if the user enters "insurance, claims, health", the system may generate different keyword combinations such as "insurance + claims", "claims + health", "insurance + health + claims", etc. To improve the rationality of the combination, the system can set limiting conditions, such as avoiding low-frequency and ineffective combinations and controlling the maximum number of combined keywords. In addition, the system can learn from the search log which keyword combinations are more valuable based on domain knowledge or big data analysis, so as to optimize the construction of search queries.

[0048] S203. Input the number of keyword combinations into the pre-trained keyword weight model in sequence to obtain the keyword weights corresponding to each keyword combination;

[0049] Specifically, the system evaluates the effectiveness of different keyword combinations through a pre-trained keyword weight model and assigns weights to them. In the specific embodiments of this application, the calculation of keyword weights can be based on multiple factors, such as the historical click-through rate (CTR) of keywords, TF-IDF word frequency statistics, semantic similarity, deep learning model scores, etc. Specific implementation methods include: Based on search logs: Analyze past search queries and user behaviors, and calculate the search popularity and user click feedback of a certain keyword combination;

[0050] Based on NLP models: Use models such as Word2Vec and BERT to calculate the semantic relevance of keywords and determine the rationality of combinations;

[0051] Based on machine learning: Use algorithms such as XGBoost and deep neural networks to predict the effectiveness of keyword combinations and generate weight scores.

[0052] The level of keyword weights directly affects subsequent sorting and screening, ensuring that more relevant keyword combinations are preferentially adopted, and improving search accuracy and efficiency.

[0053] S204. Sort the keyword combinations according to the keyword weights and determine the target keyword combination from the sorting results;

[0054] Specifically, the system will sort all the generated keyword combinations according to their weights and select the optimal combination as the final search input. During the sorting process, keyword combinations with high weights indicate better performance in past search records or better predicted search effects by the model. Sorting algorithms can use weighted scoring methods (weighted calculation considering multiple factors), ranking learning algorithms (such as LambdaMART for optimized ranking), etc. When screening target keyword combinations, the system can set certain strategies, such as selecting the top N combinations or screening based on a threshold (such as combinations with weights greater than 0.8). In addition, to further optimize the search effect, user personalized information (such as user preferences, geographical location, search history) can be combined to dynamically adjust the sorting rules, making the finally selected target keyword combinations more in line with user needs.

[0055] S205, generate a matching search query statement according to the target keyword combination;

[0056] Specifically, the system will convert the selected target keyword combination into a structured search statement that can be used for querying in the search engine. The generation method of the search query statement depends on the search platform used. For example:

[0057] SQL query: If the search is based on a database, generate SELECT / WHERE statements;

[0058] Elasticsearch / Lucene query: Use boolean queries (bool query), add AND, OR logical operators to improve search flexibility;

[0059] Natural language query conversion: Combine pre-trained models such as BERT and T5 to convert keywords into natural language search questions to adapt to intelligent search systems.

[0060] For example, the target keyword combination "insurance + claims" can be converted into "insurance AND claims" (boolean query) or "How to handle insurance claims?" (natural language query), and then submitted to the search system. In addition, to improve search accuracy, the system can add filtering conditions (such as time range, geographical restrictions), sorting rules (such as sorting by relevance or time), making the search results more in line with user needs.

[0061] S206, based on the search query statement, use a pre-trained deep learning model to perform information search and output the information search results.

[0062] Specifically, finally, the system uses a deep learning search model to perform information retrieval on the constructed query statement and outputs search results that meet the user's needs. Traditional search engines are mainly based on keyword matching, while this application introduces deep learning models (such as BERT-based search, GPT question-answering system, Dense Retrieval), which can understand search intentions and improve the relevance and accuracy of search results. The search process generally includes:

[0063] Query statement parsing: Use BERT or Transformer models to understand the query intention and optimize the query structure;

[0064] Semantic index matching: Use vectorization methods (such as FAISS, ANN) to perform semantic retrieval in a large-scale document library;

[0065] Ranking optimization: Adopt the learning to rank (LTR) method to perform relevance ranking on candidate search results and improve the click-through rate.

[0066] Finally, the system will sort the search results according to factors such as relevance, time, and weight and display them to the user. To further optimize the user experience, functions such as search result summaries, recommended related queries, and filtering conditions can also be provided to make the search more intuitive and intelligent.

[0067] In the above embodiments, this application improves the accuracy and relevance of search queries through intelligent keyword combination and weight evaluation. Compared with traditional keyword matching searches, this method is based on a pre-trained keyword weight model, which can automatically evaluate the importance of different keyword combinations, reduce invalid or low-relevance queries, and improve search efficiency. At the same time, deep learning models are used to optimize search query statements to make them more in line with user intentions and enhance the quality of search results. This not only improves the accuracy of information retrieval but also reduces the cost of manual screening and adjustment of search keywords. In a large data environment, the information method provided by this application can be applied to business management program systems such as fintech, healthcare, and elderly care to provide users with a more intelligent and efficient information search experience.

[0068] Furthermore, before the step of inputting the keyword combination sequence into the pre-trained keyword weight model to obtain the keyword weights corresponding to each keyword combination, it further includes:

[0069] Build the network architecture of the keyword weight model, where the keyword weight model includes an encoding layer, a masking layer, a self-attention layer, and a linear regression layer;

[0070] Obtain historical keyword combinations and obtain the historical information search results corresponding to the historical keyword combinations;

[0071] Iteratively train the keyword weight model based on historical keyword combinations and historical information search results.

[0072] In this embodiment, before inputting the keyword combination sequence into the pre-trained keyword weight model, it is first necessary to build the network architecture of the model to ensure that it can effectively evaluate the importance of different keyword combinations. The keyword weight model consists of an encoding layer, a masking layer, a self-attention layer, and a linear regression layer. Among them, the encoding layer is responsible for converting the input keyword combination into a vector representation, usually processed using word embeddings (Word2Vec, BERT Embedding); the masking layer is used to filter out invalid keywords or low-weight keywords to reduce noise interference; the self-attention layer (Self-Attention) is used to calculate the correlation within the keyword combination to ensure that the model focuses on high-value keyword collocations; the linear regression layer finally outputs the weight score of the keyword combination.

[0073] In the model training stage, the system needs to obtain historical keyword combinations and their corresponding search results to construct a training dataset. Historical data can come from search logs, user click records, search rankings, etc., and the system will perform supervised learning based on these data. During the training process, a loss function (such as mean squared error MSE) is used to measure the deviation between the predicted keyword weight and the actual search effect, and optimization is performed through backpropagation + gradient descent to ensure that the model can accurately predict the effectiveness of keyword combinations. After multiple rounds of iterative training, the model can be applied to real-time search tasks to improve the accuracy of keyword weight prediction.

[0074] Through the above steps, the keyword weight model can accurately evaluate the effectiveness of different keyword combinations, ensure that high-value keyword combinations are preferentially adopted during the search process, thereby improving the accuracy and relevance of information search and reducing the interference of inefficient or irrelevant search results.

[0075] Furthermore, the steps of iteratively training the keyword weight model based on historical keyword combinations and historical information search results specifically include:

[0076] Perform an encoding operation on the historical keyword combination to obtain the first keyword feature vector;

[0077] Perform a feature masking operation on the first keyword feature vector to obtain the first masked vector;

[0078] Use the self-attention mechanism to perform self-attention weighting operation on the first masked vector to obtain the first weighted feature vector;

[0079] Perform linear regression prediction on the first weighted feature vector to obtain the initial information search result;

[0080] Iteratively update the keyword weight model based on the initial information search results and the historical information search results.

[0081] In this embodiment, iteratively train the keyword weight model based on the historical keyword combinations and the historical information search results. First, it is necessary to perform an encoding operation on the historical keyword combinations to obtain the first keyword feature vector. The encoding operation usually adopts the word embedding technology to convert the keywords into high-dimensional vector representations, which is convenient for calculating the relationships between keywords. Then, through the feature masking operation, process the first keyword feature vector to remove inefficient or irrelevant keyword features, so as to enhance the model's attention to important information and generate the first masked vector.

[0082] Next, use the self-attention mechanism to perform a weighting operation on the first masked vector, enabling the model to adaptively learn the association relationships between keywords, thereby obtaining the first weighted feature vector. The self-attention mechanism can capture the dependency relationships between long-distance keywords and optimizes the feature representation ability. After that, pass the weighted feature vector through the linear regression model for prediction to obtain the initial information search results. There may be differences between the initial results and the actual historical search results. Therefore, the system compares the error between the initial search results and the historical information search results, uses the loss function for backpropagation, and adjusts the model weights to achieve the effect of iterative optimization. The model is continuously updated in each round of training to optimize its prediction ability for the keyword combination weights.

[0083] Through the above steps, the keyword weight model can be continuously iteratively optimized, improving the accurate weight prediction for keyword combinations, enhancing the intelligence and accuracy of search queries, and thus improving the quality and efficiency of information retrieval.

[0084] Furthermore, the linear regression layer includes several output ports, and each output port corresponds to the prediction result of a historical keyword. After the step of performing linear regression prediction on the first weighted feature vector to obtain the initial information search results, it further includes:

[0085] Obtain the hidden states of the output ports of each linear regression layer to get the first hidden state;

[0086] Independently score each first hidden state to obtain the feature weights of the historical keywords corresponding to each output port;

[0087] Combine the feature weights of all historical keywords to obtain the historical keyword weights corresponding to the historical keyword combinations.

[0088] In this embodiment, the system obtains the hidden states of the output ports of each linear regression layer to obtain the first hidden states. Each output port corresponds to the prediction result of a historical keyword. Therefore, the hidden state records the internal activation information of the model at this port, reflecting the performance and relevance of this keyword in the current query. Next, each first hidden state is scored independently, and by calculating its corresponding weight, the feature weight of the historical keyword corresponding to each output port is obtained. The feature weight reflects the model's evaluation of the importance of each historical keyword in the search query. The larger the weight, the more significant the role of this historical keyword in the prediction.

[0089] In this step, the system scores each first hidden state independently, that is, it evaluates the contribution value of the historical keyword corresponding to each output port separately. Specifically, the hidden state represents the internal activation information of the model at each output port, which reflects the relevance and importance of this historical keyword to the current query. Through independent scoring, the model calculates a weight value representing the influence degree of this keyword on the search results.

[0090] For example, assume that when searching for "health insurance claim", the historical keyword combination includes "health insurance" and "claim". The system will score "health insurance" and "claim" separately according to the performance of each keyword in the historical search to obtain the corresponding weights. Assume the scoring results are: the weight of "health insurance" is 0.8, and the weight of "claim" is 0.6. These weight values reflect the relative importance of each keyword in the search, helping the system to more accurately evaluate which keyword combinations can provide more relevant search results.

[0091] Finally, the feature weights of all historical keywords are combined to obtain the historical keyword weight corresponding to the final historical keyword combination. The feature weights of each keyword calculated by the model are aggregated to form a comprehensive weight distribution, representing the importance of the entire historical keyword combination in the current search task.

[0092] Through the above steps, the model can accurately evaluate the feature weights of each historical keyword in the search process, thereby forming a comprehensive historical keyword weight, optimizing the keyword combination, and further improving the intelligence and accuracy of information retrieval.

[0093] Furthermore, the steps of iteratively updating the keyword weight model based on the initial information search results and the historical information search results specifically include:

[0094] Based on the loss function of the keyword weight model, calculate the error between the initial information search results and the historical information search results to obtain the prediction error;

[0095] Use the backpropagation algorithm to transmit the prediction error through each network layer of the keyword weight model;

[0096] Adjust the network parameters of the keyword weight model until the prediction error of each network layer is less than or equal to the preset error threshold.

[0097] In this embodiment, the system calculates the error between the initial information search result and the historical information search result based on the loss function of the keyword weight model to obtain the prediction error. This error reflects the gap between the model's current prediction and the true result. Then, using the backpropagation algorithm, the prediction error is transmitted layer by layer from the output layer back to each network layer of the model to update the weights and parameters of the model. Through backpropagation, the system can identify which parameters need to be adjusted, thereby reducing the prediction error. Finally, the model is adjusted according to the preset error threshold until the prediction error of each network layer is less than or equal to the threshold. This process will be repeated until the performance of the model reaches the expectation. In this way, the system can continuously optimize to improve the performance and accuracy of the keyword weight model in actual searches.

[0098] Through the above steps, the model can self-optimize and adjust, reduce the prediction error, thereby improving the accurate evaluation of the keyword weight, and enhancing the precision and efficiency of the search query.

[0099] Further, the step of inputting several keyword combinations into the pre-trained keyword weight model in sequence to obtain the keyword weights corresponding to each keyword combination specifically includes:

[0100] Perform an encoding operation on the keyword combination to obtain a second keyword feature vector;

[0101] Perform a feature masking operation on the second keyword feature vector to obtain a second mask vector;

[0102] Use the self-attention mechanism to perform self-attention weighting operation on the second mask vector to obtain a second weighted feature vector;

[0103] Perform linear regression prediction on the second weighted feature vector;

[0104] Obtain the hidden state of the output port of each linear regression layer to get a second hidden state;

[0105] Independently score each second hidden state to obtain the feature weight of the keyword corresponding to each output port;

[0106] Combine the feature weights of all keywords to obtain the keyword weight corresponding to the keyword combination.

[0107] In this embodiment, after sequentially inputting several keyword combinations into a pre-trained keyword weight model, first, an encoding operation is performed on each keyword combination to convert each keyword into a corresponding feature vector. Usually, word embedding techniques such as Word2Vec or BERT are used to map the keywords into a high-dimensional space, obtaining the second keyword feature vector. Then, a feature masking operation is executed to filter out inefficient or irrelevant keyword features, avoiding noise interference, and generating a second mask vector, enabling the model to only focus on the keyword features useful for the search.

[0108] Then, the self-attention mechanism is used to weight the second mask vector to obtain a second weighted feature vector. The self-attention mechanism can automatically adjust the weights between keywords, enabling the keywords relevant to the query to obtain higher weights, ensuring that the model can identify and strengthen the important parts in the keyword combination and ignore the irrelevant parts. Next, through linear regression prediction, the model will calculate the predicted keyword combination result based on the second weighted feature vector.

[0109] Immediately afterwards, the system extracts the second hidden state from the output port of the linear regression layer. This is the internal activation information of the model, representing the potential features of each keyword combination. By independently scoring each second hidden state, the model can assign a feature weight to the keyword corresponding to each output port, and this weight reflects the relevance of each keyword in the search. Finally, the feature weights of all keywords are combined to obtain the total weight of the keyword combination. This total weight is the basis for the model to judge the importance and effectiveness when processing the keyword combination, optimizing the result of the search query.

[0110] Through the above steps, the keyword weight model can accurately evaluate the importance of the keyword combination, improve the efficiency and accuracy of the search engine when processing complex queries, and ensure that users obtain more relevant search results.

[0111] Furthermore, the steps of generating a matching search query statement according to the target keyword combination specifically include:

[0112] Perform entity recognition on the target keyword combination to obtain the search target entity;

[0113] According to the search target entity, identify the query statement template that matches the search target entity in the preset query statement template library;

[0114] Fill the target keyword combination into the query statement template and perform query statement syntax correction;

[0115] When the syntax correction of the query statement is completed, a search query statement that matches the target keyword combination is obtained.

[0116] In this embodiment, during the process of generating a matching search query statement according to the target keyword combination, entity recognition needs to be performed on the target keyword combination first to identify the search target entity therein. Entity recognition is an important technology in natural language processing and can identify specific entities in keywords, such as person names, place names, product names, etc. For example, when searching for "health insurance claim policy", entity recognition may identify "health insurance" as a product entity and "claim policy" as an entity related to the policy.

[0117] Next, the system searches for a query statement template that matches the target entity in the preset query statement template library according to the identified search target entity. These templates are pre-constructed based on common query patterns and structures. For example, "[entity] related information" or "policy regarding [entity]". The system will select a suitable template for filling according to the type of the target entity.

[0118] Then, the system fills the target keyword combination according to the selected template and corrects the grammar of the filled query statement. Grammar correction ensures that the query statement conforms to grammar rules and has a fluent and natural language expression, enabling the search engine to better understand the query intention and perform retrieval operations. After completing the grammar correction, the system can generate a search query statement that highly matches the target keyword combination.

[0119] Through the above steps, the generated search query statement is more in line with the user's search intention and natural language expression, improving the relevance and accuracy of the search and enhancing the understanding ability of the search engine.

[0120] In the above embodiment, the present application discloses an information search method, belonging to the field of artificial intelligence technology. Through intelligent keyword combination and weight evaluation, the present application improves the accuracy and relevance of search queries. Compared with traditional keyword matching searches, this method is based on a pre-trained keyword weight model and can automatically evaluate the importance of different keyword combinations, reducing invalid or low-relevance queries and improving search efficiency. At the same time, a deep learning model is used to optimize the search query statement to make it more in line with the user's intention and enhance the quality of search results. It not only improves the accuracy of information retrieval but also reduces the cost of manual screening and adjustment of search keywords. In a massive data environment, the information method provided by the present application can be applied to business management program systems such as fintech, healthcare, and elderly care to provide users with a more intelligent and efficient information search experience.

[0121] In this embodiment, the electronic device on which the information search method runs (for example Figure 1The server shown can receive instructions or obtain data through a wired connection or a wireless connection. It should be noted that the above wireless connection methods can include, but are not limited to, 3G / 4G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra wideband) connections, and other currently known or future-developed wireless connection methods.

[0122] It should be emphasized that to further ensure the privacy and security of the above information search result information, the above information search result information can also be stored in a node of a blockchain.

[0123] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithms. A blockchain, in essence, is a decentralized database, a string of data blocks generated by using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer, etc.

[0124] The embodiments of this application can obtain and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use the knowledge to obtain the best results of theory, methods, technologies, and application systems.

[0125] Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, robotics, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0126] Those of ordinary skill in the art can understand that all or part of the processes in implementing the above method embodiments can be completed by instructing relevant hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When this program is executed, it can include the processes of the above method embodiments. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, an optical disc, a read-only memory (ROM), or a random access memory (RAM), etc.

[0127] It should be understood that although the steps in the flowchart of the accompanying drawings are shown sequentially according to the indication of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless otherwise clearly stated in this document, there is no strict order restriction for the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps.

[0128] Further reference Figure 3 , as an implementation of the method shown above Figure 2 , this application provides an embodiment of an information search device. This device embodiment corresponds to Figure 2 the method embodiment shown, and this device can be specifically applied to various electronic devices.

[0129] As Figure 3 shown, the information search device 300 described in this embodiment includes:

[0130] A keyword acquisition module 301, configured to receive an information search instruction and acquire a sequence of search keywords uploaded by a user, where the sequence of search keywords includes a plurality of search keywords;

[0131] A keyword combination module 302, configured to randomly combine the search keywords in the sequence of search keywords to obtain a plurality of keyword combinations;

[0132] A weight generation module 303, configured to sequentially input the plurality of keyword combinations into a pre-trained keyword weight model to obtain the keyword weights corresponding to each keyword combination;

[0133] A weight sorting module 304, configured to sort the keyword combinations according to the keyword weights and determine a target keyword combination from the sorting result;

[0134] A statement generation module 305, configured to generate a matching search query statement according to the target keyword combination;

[0135] An information search module 306, configured to perform information search based on the search query statement using a pre-trained deep learning model and output an information search result.

[0136] Furthermore, the information search device 300 further includes:

[0137] A model construction module for constructing the network architecture of a keyword weight model, where the keyword weight model includes an encoding layer, a masking layer, a self-attention layer, and a linear regression layer;

[0138] A historical keyword module for obtaining historical keyword combinations and obtaining historical information search results corresponding to the historical keyword combinations;

[0139] An iterative training module for iteratively training the keyword weight model based on the historical keyword combinations and the historical information search results.

[0140] Furthermore, the iterative training module specifically includes:

[0141] A first encoding unit for encoding the historical keyword combination to obtain a first keyword feature vector;

[0142] A first masking unit for performing a feature masking operation on the first keyword feature vector to obtain a first masked vector;

[0143] A first weighting unit for performing a self-attention weighting operation on the first masked vector using the self-attention mechanism to obtain a first weighted feature vector;

[0144] A first regression prediction unit for performing a linear regression prediction on the first weighted feature vector to obtain an initial information search result;

[0145] An iterative training unit for iteratively updating the keyword weight model based on the initial information search result and the historical information search result.

[0146] Furthermore, the linear regression layer includes a number of output ports, and each output port corresponds to a prediction result of a historical keyword. The iterative training module further includes:

[0147] A first hidden unit for obtaining the hidden states of the output ports of each linear regression layer to obtain a first hidden state;

[0148] A first scoring unit for independently scoring each first hidden state to obtain the feature weights of the historical keywords corresponding to each output port;

[0149] A first weight calculation unit for combining the feature weights of all historical keywords to obtain the historical keyword weights corresponding to the historical keyword combination.

[0150] Furthermore, the iterative training unit specifically includes:

[0151] A prediction error sub-unit for calculating the error between the initial information search result and the historical information search result based on the loss function of the keyword weight model to obtain a prediction error;

[0152] An error propagation subunit, configured to use the backpropagation algorithm to propagate the prediction error through each network layer of the keyword weight model;

[0153] An iterative training subunit, configured to adjust the network parameters of the keyword weight model until the prediction error of each network layer is less than or equal to a preset error threshold.

[0154] Furthermore, the weight generation module 303 specifically includes:

[0155] A second encoding unit, configured to perform an encoding operation on the keyword combination to obtain a second keyword feature vector;

[0156] A second masking unit, configured to perform a feature masking operation on the second keyword feature vector to obtain a second mask vector;

[0157] A second weighting unit, configured to perform a self-attention weighting operation on the second mask vector using the self-attention mechanism to obtain a second weighted feature vector;

[0158] A second regression prediction unit, configured to perform a linear regression prediction on the second weighted feature vector;

[0159] A second hidden unit, configured to obtain the hidden states of the output ports of each linear regression layer to obtain a second hidden state;

[0160] A second scoring unit, configured to independently score each second hidden state to obtain the feature weights of the keywords corresponding to each output port;

[0161] A second weight calculation unit, configured to combine the feature weights of all keywords to obtain the keyword weight corresponding to the keyword combination.

[0162] Furthermore, the statement generation module 305 specifically includes:

[0163] An entity recognition unit, configured to perform entity recognition on the target keyword combination to obtain a search target entity;

[0164] A template matching unit, configured to identify a query statement template that matches the search target entity in a preset query statement template library according to the search target entity;

[0165] A statement processing unit, configured to fill the target keyword combination into the query statement template and perform query statement syntax correction;

[0166] A statement output unit, configured to obtain a search query statement that matches the target keyword combination when the query statement syntax correction is completed.

[0167] In the above embodiments, the present application discloses an information search device, belonging to the field of artificial intelligence technology. Through intelligent keyword combination and weight evaluation, the present application improves the accuracy and relevance of search queries. Compared with traditional keyword matching searches, this method is based on a pre-trained keyword weight model, which can automatically evaluate the importance of different keyword combinations, reduce invalid or low-relevance queries, and improve search efficiency. At the same time, a deep learning model is used to optimize the search query statement to make it more in line with the user's intention, enhance the quality of search results, not only improve the accuracy of information retrieval, but also reduce the cost of manual screening and adjustment of search keywords. In a massive data environment, the information method provided by the present application can be applied to business management program systems such as fintech, healthcare and elderly care, etc., to provide users with a more intelligent and efficient information search experience.

[0168] To solve the above technical problems, the embodiments of the present application also provide a computer device. For details, please refer to Figure 4 , Figure 4 which is the basic structural block diagram of the computer device in this embodiment.

[0169] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are communicatively connected to each other through a system bus. It should be noted that only the computer device 4 with a memory 41, a processor 42, and a network interface 43 is shown in the figure, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented. Among them, those skilled in the art of the present technology can understand that a computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0170] The computer device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The computer device can perform human-computer interaction with the user through a keyboard, a mouse, a remote control, a touchpad, a voice control device, etc.

[0171] The memory 41 at least includes one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a Smart Media Card (SMC), a Secure Digital (SD) card, a FlashCard, etc. Of course, the memory 41 may also include both the internal storage unit and the external storage device of the computer device 4. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the information search method. In addition, the memory 41 may also be used to temporarily store various data that have been output or will be output.

[0172] In some embodiments, the processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run the computer-readable instructions stored in the memory 41 or process data, such as running the computer-readable instructions of the information search method.

[0173] The network interface 43 may include a wireless network interface or a wired network interface, and the network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0174] An information search method includes:

[0175] Receiving an information search instruction, and obtaining a search keyword sequence uploaded by a user, where the search keyword sequence includes a plurality of search keywords;

[0176] Randomly combining the search keywords in the search keyword sequence to obtain a plurality of keyword combinations;

[0177] Sequentially inputting the plurality of keyword combinations into a pre-trained keyword weight model to obtain keyword weights corresponding to the respective keyword combinations;

[0178] Sort the keyword combinations according to the keyword weights, and determine the target keyword combination from the sorting result;

[0179] Generate a matching search query statement according to the target keyword combination;

[0180] Based on the search query statement, use a pre-trained deep learning model to perform information search and output the information search result.

[0181] This application also provides another implementation manner, that is, to provide a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor, so that the at least one processor executes the steps of the information search method as described above.

[0182] An information search method includes:

[0183] Receive an information search instruction, and obtain a search keyword sequence uploaded by a user, where the search keyword sequence includes several search keywords;

[0184] Randomly combine the search keywords in the search keyword sequence to obtain several keyword combinations;

[0185] Input the several keyword combinations into a pre-trained keyword weight model in sequence to obtain the keyword weights corresponding to each keyword combination;

[0186] Sort the keyword combinations according to the keyword weights, and determine the target keyword combination from the sorting result;

[0187] Generate a matching search query statement according to the target keyword combination;

[0188] Based on the search query statement, use a pre-trained deep learning model to perform information search and output the information search result.

[0189] Through the description of the above implementation manners, those skilled in the art can clearly understand that the method of the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation manner. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc), and includes several instructions for causing a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in the various embodiments of this application.

[0190] This application can be used in numerous general-purpose or special-purpose computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This application can also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0191] It should be noted that the non-company software tools or components appearing in each embodiment of this application are only introduced by way of example and do not represent actual use.

[0192] Obviously, the embodiments described above are only a part of the embodiments of this application, rather than all of them. The preferred embodiments of this application are given in the drawings, but do not limit the patent scope of this application. This application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of this application more thorough and comprehensive. Although this application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions described in the foregoing specific embodiments, or perform equivalent replacements for some of the technical features. Any equivalent structures directly or indirectly using the content of this application's specification and drawings in other related technical fields are equally within the scope of the patent protection of this application.

Claims

1. An information search method, characterized in that: include: Receiving an information search instruction, obtaining a search keyword sequence uploaded by a user, wherein the search keyword sequence includes a plurality of search keywords; Randomly combining the search keywords in the search keyword sequence to obtain a plurality of keyword combinations; Inputting a plurality of the keyword combinations into a pre-trained keyword weight model in sequence to obtain a keyword weight corresponding to each of the keyword combinations; Sorting the keyword combinations according to the keyword weights, and determining a target keyword combination from the sorting results; Generating a matching search query statement according to the target keyword combination; Based on the search query statement, a pre-trained deep learning model is used to perform information search and output information search results.

2. The information search method according to claim 1, characterized in that: Before the step of inputting the keyword combination sequence into the pre-trained keyword weight model to obtain the keyword weight corresponding to each keyword combination, the method further includes: Building a network architecture of the keyword weight model, wherein the keyword weight model includes an encoding layer, a mask layer, a self-attention layer, and a linear regression layer; Obtaining a historical keyword combination, and obtaining a historical information search result corresponding to the historical keyword combination; The keyword weight model is iteratively trained based on the historical keyword combination and the historical information search results.

3. The information search method according to claim 2, characterized in that: The step of iteratively training the keyword weight model based on the historical keyword combination and the historical information search results specifically includes: Performing an encoding operation on the historical keyword combination to obtain a first keyword feature vector; Performing a feature mask operation on the first keyword feature vector to obtain a first mask vector; Using a self-attention mechanism to perform a self-attention weighted operation on the first mask vector to obtain a first weighted feature vector; Performing linear regression prediction on the first weighted feature vector to obtain initial information search results; The keyword weight model is iteratively updated based on the initial information search results and the historical information search results.

4. The information search method according to claim 3, characterized in that: The linear regression layer includes a plurality of output ports, each of which corresponds to a prediction result of a historical keyword. After the step of performing linear regression prediction on the first weighted feature vector to obtain the initial information search result, the method further includes: Obtaining the hidden state of the output port of each of the linear regression layers to obtain a first hidden state; Scoring each of the first hidden states independently to obtain a feature weight of a historical keyword corresponding to each of the output ports; The feature weights of all historical keywords are combined to obtain the historical keyword weight corresponding to the historical keyword combination.

5. The information search method according to claim 3, characterized in that: The step of iteratively updating the keyword weight model based on the initial information search results and the historical information search results specifically includes: Based on the loss function of the keyword weight model, calculating the error between the initial information search result and the historical information search result to obtain a prediction error; Using a back propagation algorithm to propagate the prediction error in each network layer of the keyword weight model; The network parameters of the keyword weight model are adjusted until the prediction error of each network layer is less than or equal to a preset error threshold.

6. The information search method according to claim 1, characterized in that: The step of sequentially inputting a plurality of keyword combinations into a pre-trained keyword weight model to obtain keyword weights corresponding to each keyword combination specifically includes: Performing an encoding operation on the keyword combination to obtain a second keyword feature vector; Performing a feature mask operation on the second keyword feature vector to obtain a second mask vector; Using a self-attention mechanism to perform a self-attention weighted operation on the second mask vector to obtain a second weighted feature vector; Performing linear regression prediction on the second weighted feature vector; Obtaining the hidden state of the output port of each of the linear regression layers to obtain a second hidden state; Scoring each of the second hidden states independently to obtain a feature weight of a keyword corresponding to each of the output ports; The feature weights of all keywords are combined to obtain the keyword weight corresponding to the keyword combination.

7. The information search method according to claim 1, characterized in that: The step of generating a matching search query statement according to the target keyword combination specifically includes: Performing entity recognition on the target keyword combination to obtain a search target entity; According to the search target entity, identifying a query statement template matching the search target entity in a preset query statement template library; Filling the target keyword combination into the query statement template and performing query statement grammar correction; After the grammatical correction of the query statement is completed, a search query statement matching the target keyword combination is obtained.

8. An information search device, characterized in that: include: A keyword acquisition module, used to receive an information search instruction and acquire a search keyword sequence uploaded by a user, wherein the search keyword sequence includes a plurality of search keywords; A keyword combination module, used for randomly combining the search keywords in the search keyword sequence to obtain a plurality of keyword combinations; A weight generation module, used to input a plurality of the keyword combinations into a pre-trained keyword weight model in sequence to obtain a keyword weight corresponding to each of the keyword combinations; A weight sorting module, used to sort the keyword combinations according to the keyword weights, and determine a target keyword combination from the sorting results; A statement generation module, used to generate a matching search query statement according to the target keyword combination; The information search module is used to perform information search based on the search query statement using a pre-trained deep learning model and output information search results.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the information search method according to any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the information search method according to any one of claims 1 to 7 are implemented.

Images