Document retrieval

Abstract

The present disclosure relates to the technical field of artificial intelligence, and provides a document retrieval method and apparatus, an electronic device, a storage medium, and a product. The method comprises: in response to a document retrieval request sent by a user account and comprising target retrieval text, processing the target retrieval text to obtain a retrieval word in a preset format, wherein the preset format is a data format which can be identified by a preset retrieval interface, and the preset retrieval interface is an interface capable of using a plurality of retrieval paths to perform retrieval; on the basis of the retrieval word and the preset retrieval interface, using the plurality of retrieval paths to perform retrieval to obtain a document retrieval result corresponding to each retrieval path; and sending the document retrieval results corresponding to the plurality of retrieval paths to the user account.

Description

[0001] Document retrieval technology field

[0002]

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to document retrieval.

[0003] With the rapid development of information technology, service systems have evolved into highly complex and feature-rich network structures, which can be composed of multi-layered data architectures, numerous subsystems, and cross-platform integrated modules. The number of documents in service systems is very large, and a certain method is required for retrieval.

[0004]

[0003] In the related art, a search text input by a user account is obtained, keywords are extracted from the search text, and then based on the extracted keywords, a query is made to see whether the keyword is included in the inverted index of the search engine. If the keyword is included, the document corresponding to the keyword is returned to the user account.

[0005]

[0004] However, the related art only performs retrieval based on keywords extracted from the retrieval text, and the retrieval results are not accurate enough.

[0006]

[0005] Embodiments of the present disclosure provide a document retrieval method, apparatus, electronic device, storage medium, and product that can employ multiple retrieval paths to perform retrieval and obtain more accurate retrieval results. The technical solution is as follows.

[0007]

[0006] In a first aspect, a document retrieval method is provided, the method comprising: in response to a document retrieval request containing a target retrieval text sent by a user account, processing the target retrieval text to obtain a retrieval term in a preset format, wherein the preset format is a data format recognizable by a preset retrieval interface, and the preset retrieval interface is an interface capable of performing retrieval using multiple retrieval paths; based on the retrieval term and the preset retrieval interface, performing retrieval using multiple retrieval paths to obtain document retrieval results corresponding to each retrieval path; and sending the document retrieval results corresponding to the multiple retrieval paths to the user account.

[0008]

[0007] In a second aspect, a document retrieval device is provided, the device comprising: a processing module for processing the target retrieval text in response to a document retrieval request sent by a user account and containing target retrieval text to obtain a search term in a preset format, wherein the preset format is a data format recognizable by a preset retrieval interface, wherein the preset retrieval interface is an interface capable of performing retrieval using multiple retrieval paths; a retrieval module for performing retrieval using multiple retrieval paths based on the retrieval term and the preset retrieval interface, and obtaining document retrieval results corresponding to each retrieval path; and a sending module for sending the document retrieval results corresponding to the multiple retrieval paths to the user account.

[0008] In a third aspect, an electronic device is provided, comprising a processor and a memory; the memory storing at least one program code; the at least one program code being configured to be called and executed by the processor to implement the document retrieval method described in the first aspect.

[0009]

[0009] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores at least one computer program, and when the at least one computer program is executed by a processor, it can implement the document retrieval method as described in the first aspect.

[0010]

[0010] In a fifth aspect, a computer program product is provided, wherein the computer program product includes a computer program, and when the computer program is executed by a processor, the document retrieval method as described in the first aspect can be implemented.

[0011]

[0011] The beneficial effects of the technical solution provided by the embodiment of the present disclosure are as follows: The embodiment of the present disclosure provides a preset search interface, which can use multiple search paths for searching. When a document search request carrying a target search text is received from a user account, the target search text is processed to obtain a search term in a preset format. The preset format is a data format that can be recognized by the preset search interface. Therefore, the preset search interface can use multiple search paths to search based on the search term, thereby obtaining document search results corresponding to each search path. Compared with searching based on keywords extracted from the search text, the search method is richer and the search results are more accurate.

[0012]

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0013]

[0013] FIG1 is a schematic diagram of an implementation environment involved in a document retrieval method provided in an embodiment of the present disclosure;

[0014] FIG. 2 is a schematic diagram of a document retrieval system according to an embodiment of the present disclosure;

[0015]

[0015] FIG3 is a schematic diagram of a method for constructing an index in a document retrieval system provided in an embodiment of the present disclosure;

[0016] FIG4 is a flow chart of a document retrieval method according to an embodiment of the present disclosure;

[0017] FIG5 is a schematic diagram of a document retrieval process provided by an embodiment of the present disclosure;

[0018]

[0018] FIG6 is a schematic structural diagram of a document retrieval device provided in an embodiment of the present disclosure;

[0019]

[0019] FIG7 shows a block diagram of an electronic device according to an exemplary embodiment of the present disclosure.

[0020]

[0020] In order to make the purpose, technical solutions and advantages of the present disclosure more clear, the embodiments of the present disclosure will be further described in detail below with reference to the accompanying drawings.

[0021]

[0021] It is understood that the terms "each," "a plurality," and "any" used in the embodiments of the present disclosure include two or more, each refers to each of the corresponding plurality, and any refers to any one of the corresponding plurality. For example, if a plurality of words includes 10 words, each word refers to each of the 10 words, and any word refers to any one of the 10 words.

[0022]

[0022] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0023]

[0023] Before executing the embodiments of the present disclosure, the terms involved in the embodiments of the present disclosure are first explained.

[0024]

[0024] Large Language Models (LLMs) are a type of complex algorithmic model built using deep learning technology that can process, understand, and generate natural language text. Large language models are usually based on transformers.

[0025] (Transformer) architecture, which is a self-attention mechanism that can capture long-range dependencies in input text sequences.

[0026] A retrieval system is an information system that helps users find and retrieve specific information from large, complex data sets. The primary purpose of a retrieval system is to provide an efficient method for users to find desired data or documents based on various search criteria. Retrieval systems are widely used in areas such as internet search engines, library catalogs, corporate databases, electronic archives management, and professional knowledge bases.

[0027]

[0026] Vectorization refers to the process of converting data structures into vector form in the fields of computer science and natural language processing. This process is particularly important in machine learning and data analysis because most algorithmic models are based on mathematical operations, which are usually performed in vector space. In short, vectorization is a way of representing data using numerical vectors.

[0028]

[0027] A search engine is an online information retrieval system designed to help users find content on the Internet by entering query keywords or phrases. Search engines crawl, index, and rank web pages to provide fast, relevant search results. These results are usually displayed in the form of a list of links, each link pointing to a web page and including a title and a brief description or web page summary.

[0028] An inverted index is a database indexing technology widely used in full-text search engines. In an inverted index, the index is built around content keywords, with the goal of quickly locating documents containing specific keywords. This indexing method is in contrast to traditional forward indexing, which organizes data by document and records the attributes and content of each document.

[0029] Natural Language Processing (NLP) is an interdisciplinary field in computer science, artificial intelligence, and linguistics. Its goal is to develop algorithms and techniques that enable computers to understand, interpret, and generate content in human language. NLP involves the analysis and processing of basic language units (such as words, phrases, and sentences) to complex text and speech conversations.

[0030] Artificial intelligence (AI) refers to theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also involves studying the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0031]

[0031] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, autonomous driving, drones, robots, smart medical care, and smart customer service. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role. The solutions provided in the embodiments of this disclosure involve artificial intelligence technologies such as natural language processing, which will be specifically explained through the subsequent embodiments. Natural language processing is an important field in the fields of computer science and artificial intelligence. It studies various theories and methods that enable effective communication between humans and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language, that is, the language used by people in daily life, and is closely related to the study of linguistics. Natural language processing technology generally includes text processing, semantic understanding, machine translation, robot question answering, knowledge graphs, and other technologies.

[0032] With the rapid development of information technology, service systems have evolved into highly complex and feature-rich network structures. These systems are often composed of multi-level data architectures, numerous subsystems, and cross-platform integrated modules to meet the diverse needs of service systems for data, entities, and functions. Taking the log system as an example, the data, entities, and functions in the log system are huge in number, requiring a large amount of time and energy to use. In addition, certain professional and experience restrictions are required, and accessing each function requires a specific operation path, which greatly reduces the user experience. At the same time, due to the emergence of large language models, human-computer interaction methods are also changing rapidly. Therefore, how to find a more convenient and efficient text retrieval method that can support both humans and large language models has become a problem of great concern to those skilled in the art.

[0033]

[0033] The disclosed embodiments provide a document retrieval method. The input retrieval text of the method can be specific data within a service system, category data within the service system, or semantically complete natural language or semantic fragments. Regardless of which of the above is the input retrieval text, the method can intelligently retrieve corresponding information. This information comes from multiple aspects of the service system and has characteristics such as diversity, complexity, and expert experience. Compared with retrieval methods based on keywords extracted from the retrieval text, the retrieved information is richer and more comprehensive. Furthermore, the disclosed embodiments do not require the extensive time and cost of training a large language model. Instead, the method simply fine-tunes the customized output of the large language model to a specified interface, which then directs the search to different search paths. This not only achieves an efficient, low-cost, and easy-to-understand retrieval method, but also provides a new human-computer interaction paradigm.

[0034]

[0034] Please refer to FIG. 1 , which shows an implementation environment involved in the document retrieval method provided by an embodiment of the present disclosure. Referring to FIG. 1 , the implementation environment includes a terminal 101 and a server 102. The terminal 101 and the server 102 can communicate directly or indirectly via a network 103, which can be a wired network or a wireless network.

[0035]

[0035] The terminal 101 may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal 101 may have an application for retrieving documents installed therein. Based on the installed application, the terminal 101 may send a document retrieval request carrying a target retrieval text to the server 102, so that the server 102 may retrieve documents that meet the user's requirements and then return the retrieved documents to the user.

[0036]

[0036] The server 102 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It may also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, and big data and artificial intelligence platforms. The server 102 may receive a document search request carrying a target search text sent by the terminal 101, process the target search text to obtain a search term in a preset format, and then, based on the search term and a preset search interface, perform a search using multiple search paths to obtain document search results corresponding to each search path. The server 102 may then send the document search results corresponding to the multiple search paths to the user account.

[0037]

[0037] FIG2 shows a framework diagram of an information fusion retrieval system provided by an embodiment of the present disclosure. Referring to FIG2 , the information fusion retrieval system includes two main layers, namely a unified input layer and a multi-path retrieval fusion layer.

[0038]

[0038] The unified input layer primarily inputs three types of search text: the first type is specific data recognizable within the service system, such as entity data; the second type is categorical data recognizable within the service system; and the third type is natural language. Specific data is primarily used to search document databases or documents from specific sources. When the search text input is specific data, the system automatically inputs it into the information fusion search interface (i.e., the preset search interface described in the embodiments of this disclosure) to quickly retrieve the required documents. Categorical data is primarily used to search for web pages with specific functions within the system. When the search text input is categorical data, the system performs further analysis and extraction to determine the type of the search text, then converts it into a pre-formatted search term, which is then input into the information fusion search interface for retrieval. When the search text input is natural language unrecognizable by the service system, the system invokes a large language model and, through interaction with the user, determines the user's intent and needs. Based on the user's intent and needs, the system then generates a pre-formatted search term, which is then input into the information fusion search interface for retrieval.

[0039]

[0039] The multi-path search fusion layer includes a multiple search path module and a unified result ranking module. The multiple search path module is used to provide multiple search paths. By using multiple search paths, the system can quickly obtain search results by performing parallel searches on the search terms input into the information fusion search interface. The multiple search paths include document search paths, vector search paths, and internal interface search paths. The document search path directly inputs the search terms, the vector search path inputs feature vectors generated based on the search terms, and the internal interface search path inputs interface data generated based on the search terms. For the document search path, the system can directly input the search terms into the document retrieval system. The document retrieval system queries the document database to see if there are document keywords matching the search terms. If the document database contains document keywords matching the search terms, the system returns the documents corresponding to the matching document keywords. The document search path fully utilizes the system's search capabilities to quickly and accurately complete information search and matching. For the vector search path, the system generates a corresponding feature vector based on the search term. This feature vector is then input into the vector search system. The vector search system queries the vector database for feature vectors with a similarity greater than a preset threshold. If a feature vector with a similarity greater than the preset threshold is stored in the vector database, the document corresponding to the feature vector with similarity greater than the preset threshold is returned. This vector search path leverages the efficiency of vector computation, enabling the system to achieve higher accuracy and speed in information retrieval. For the internal interface search path, the system converts the search term into the data format of the internal data query interface, then calls the internal system interface to retrieve documents from the database corresponding to the corresponding service interface. Depending on the service scenario, multiple links calling the internal system interface may exist simultaneously. For documents returned by calling different interfaces, the overlap ratio needs to be evaluated and queried. This internal interface search path leverages the efficiency of the internal system interface, providing the system with better support for handling large-scale data and complex service scenarios. The unified result ranking module is used to uniformly rank and merge the results retrieved by multiple search paths, and then provide the ranked and merged results to the user. This process can be implemented through algorithmic models, ensuring the comprehensiveness, accuracy, and real-time nature of search results. Through this series of operations, the information fusion retrieval system can provide users with efficient, accurate, and convenient information query services.

[0040] FIG3 illustrates the construction method of the retrieval system index involved in the information fusion retrieval system. This index comprehensively utilizes data from multiple sources, including documents based on entity generalizations and abstractions (such as service pages), service static documents, and documents extracted from service data. This index data is of vital importance to the information fusion retrieval system and can effectively improve the efficiency and accuracy of the information fusion retrieval system. Furthermore, the information fusion retrieval system may also involve some unstructured data. To further optimize the retrieval system's capabilities, multimodal vector embedding tools can be used to convert this unstructured data into vector form and store it in a vector database, thereby supporting more diverse retrieval methods and providing users with a better retrieval experience.

[0041]

[0041] The disclosed embodiments combine deep learning and natural language processing technologies to provide a new solution for complex human-computer interaction systems. This method introduces a large-scale language model to identify the intent of input natural language and convert it into input data in a preset format, thereby achieving effective natural language retrieval. In addition, the disclosed embodiments also utilize text vectorization tools to model and extract information in the service system to achieve an indexable structure. In terms of retrieval, a document retrieval system and inverted index technology can be used to quickly retrieve the abstracted information and interact with the functional module interface within the service system, thereby achieving unified sorting of multiple data. This approach not only increases the freedom of input and human-computer interaction, but also improves the data management efficiency and query performance of the service system.

[0042]

[0042] The embodiment of the present disclosure provides a document retrieval method. Referring to FIG4 , the method flow provided by the embodiment of the present disclosure includes the following steps.

[0043]

[0043] 401> In response to a document retrieval request including a target retrieval text sent by a user account, the target retrieval text is processed to obtain a retrieval term in a preset format.

[0044]

[0044] The target search text may be specific data recognizable within the service system, categorical data recognizable within the service system, or natural language not recognizable by the service system. The preset format is a data format recognizable by a preset search interface, and the preset format may include specified parameters, such as "qurey" and "type." The preset search interface is an interface capable of performing searches using multiple search paths, and may be, for example, the information fusion search interface shown in FIG. 2 . The search term may represent the search intent of the user account.

[0045]

[0045] In an embodiment of the present disclosure, in response to a document retrieval request containing a target retrieval text sent by a user account, the target retrieval text is processed to obtain a retrieval term in a preset format, and the following method can be used.

[0046]

[0046] 4011. In response to a document retrieval request, identify text content attributes of a target retrieval text.

[0047]

[0047] The text content attributes of the target search text include specific data identifiable within the service system, category data identifiable within the service system, and natural language. In response to a document search request, a content recognition model can be called to recognize the target search text, thereby identifying the text content attributes of the target search text.

[0048]

[0048] 4012. According to the text content attributes of the target search text, different processing methods are adopted to process the target search text to obtain search terms.

[0049] The text content attributes of the target retrieval text include but are not limited to the following situations.

[0049]

[0050] In the first case, when the text content attribute of the target search text is specific data or category data within the system, the target search text can be processed by word segmentation and string matching to obtain the search terms.

[0050]

[0051] In the second case, when the text content attribute of the target search text is natural language, the large language model can be called to recognize the target search text to obtain the search terms.

[0051]

[0052] The large language model is called to recognize the target search text to obtain the search terms, which may specifically include the following steps.

[0052]

[0053] The first step is to call the large language model to identify the target retrieval text and obtain multiple candidate intent information.

[0053]

[0054] For example, the content of the target search text is "I want to query the host". By calling the large language model to recognize the target search text, multiple candidate intent information can be obtained, such as "Do you want to query the information of the host?" "Do you want to query the parameters of the motherboard?", etc., and then the identified multiple candidate intent information are provided to the user account.

[0054]

[0055] The second step is to obtain the candidate intent information selected by the user account in response to the user account's selection operation on any candidate intent information.

[0055]

[0056] After receiving multiple candidate intent information identified by the large language model, the terminal displays the multiple candidate intent information. The user account can select any candidate intent information from the multiple candidate intent information based on its actual query intention. When the user account's selection operation on any candidate intent information is detected, the terminal obtains the candidate intent information selected by the user account, and then sends the candidate intent information to the server. When receiving the candidate intent information selected by the user account sent by the terminal, the server obtains the candidate intent information.

[0056]

[0057] The third step is to process the candidate intent information to obtain the search terms.

[0057]

[0058] The server can invoke the large language model to process the candidate intent information selected by the user account and obtain the search term corresponding to the candidate intent information. For example, if the candidate intent information selected by the user account is "Do you want to query host information?", the server can invoke the large language model to process the candidate intent information and obtain the search term "host information."

[0058]

[0059] The embodiments of the present disclosure provide a new human-computer interaction paradigm. By leveraging a large language model and interacting with the user, the user's intentions and needs can be identified, thereby enabling effective retrieval of natural language and improving the freedom of input and human-computer interaction.

[0059]

[0060] By processing the target search text in the document search request, a search term in a unified format can be obtained. The search term can be read by a preset search interface, so that a search can be performed using multiple search paths based on the search term.

[0060]

[0061] 402. Based on the search terms and the preset search interface, multiple search paths are used to perform searches, and document search results corresponding to each search path are obtained.

[0061]

[0062] The search path is the link used in the search process. Different links correspond to different search methods, thereby retrieving different information and enriching the search results. Specifically, based on the search terms and the preset search interface, multiple search paths are used to perform searches, and the document search results corresponding to each search path can be obtained using the following method.

[0062]

[0063] 4021. Based on the preset search interface, the search terms are processed to obtain the search path feature data corresponding to different search paths.

[0063]

[0064] Because the embodiments of the present disclosure provide multiple search paths, different search path feature data is generated for searches using different search paths. This search path feature data is the data required for searches using the corresponding search paths. Therefore, for each of the multiple search paths, the server processes the search terms based on a preset search interface, including but not limited to the following situations.

[0064]

[0065] In the first case, for the document retrieval path, the search term can be directly used as the retrieval path feature data corresponding to the document retrieval path based on the preset retrieval interface.

[0065]

[0066] In the second case, for a vector search path, a target feature vector corresponding to the search term can be obtained based on a preset search interface, and the target feature vector can be used as the search path feature data corresponding to the vector search path. When obtaining the target feature vector corresponding to the search term, a model for extracting feature vectors from the search term can be pre-trained. By invoking this model to process the search term, the target feature vector corresponding to the search term can be obtained.

[0066]

[0067] In the third case, for an internal interface search path, the search term can be converted into interface data recognizable within the system based on a preset search interface, and the interface data can then be used as the search path feature data corresponding to the internal interface search path. When converting the search term into interface data recognizable within the system, a correspondence between the search term and the interface data can be pre-set, and the search term can then be converted into interface data recognizable within the system based on the set correspondence.

[0067]

[0068] 4022. Perform a search based on the search path feature data corresponding to different search paths to obtain a document search result corresponding to each search path.

[0068]

[0069] For the above three search paths, in this step, the server performs searches based on the search path feature data corresponding to different search paths, including but not limited to the following three cases. The first case is that the search path is a document search path.

[0069]

[0070] In this case, based on the search term, a document database can be searched for document keywords that match the search term. The document database uses an inverted document index to store the correspondence between document keywords and documents. If a document keyword matching the search term is found in the document database, the document corresponding to the matching document keyword is used as the document search result corresponding to the document search path.

[0070]

[0071] The document search path can be used to retrieve documents containing document keywords, meeting the user's demand for accurate document search. The second case is that the search path is a vector search path.

[0071]

[0072] For the second case, based on the target feature vector, a vector database can be searched for feature vectors whose similarity to the target feature vector exceeds a preset threshold. The vector database stores the correspondence between feature vectors and documents. If the vector database contains a feature vector whose similarity to the target feature vector exceeds the preset threshold, the document corresponding to the found feature vector is used as the document retrieval result for the vector search path.

[0072]

[0073] Using the vector search path can retrieve documents with high similarity to the target feature vector, meeting the user's demand for fast document query. The third case is that the search path is the internal interface search path.

[0073]

[0074] In this case, based on the interface data, the corresponding service interface within the system can be called for search, obtaining document search results corresponding to the internal interface search path. Typically, the number of service interfaces that can be called by the interface data can be one or more, depending on the actual search requirements. Different service interfaces can correspond to different service scenarios, enabling the query of data generated by different service scenarios within the service system.

[0074]

[0075] Using the internal interface search path for searching can retrieve data from different service scenarios, making the search results more comprehensive and accurate.

[0075]

[0076] In this embodiment, the service system is characterized by diversity, complexity, and expert experience. This modeling creates an indexable structure for the service system. This allows for rapid retrieval of this abstracted information using text vectorization tools and a search system using an inverted index. Simultaneously, the extracted information is queried through the functional module interfaces within the service system.

[0076]

[0077] 403. Send document retrieval results corresponding to the multiple search paths to the user account.

[0077]

[0078] Document retrieval results from multiple search paths can be sent directly to the user account. Alternatively, the document retrieval results from multiple search paths can be merged to generate a merged result, which can then be sent to the user account. When merging document retrieval results from multiple search paths, duplicates must be removed and sorted, and then the results are sent to the user account based on the sorting. By uniformly sorting the results returned from multiple search paths, better retrieval results are achieved.

[0079] Existing retrieval systems primarily search based on keywords extracted from search text and are unable to summarize and extract information from the search text. Furthermore, existing retrieval systems only utilize inverted document technology for retrieval, lacking the ability to search data from other paths within the system, resulting in an incomplete search. Furthermore, existing retrieval systems are unable to adapt to the highly flexible nature of natural language input, resulting in a relatively limited search process. Furthermore, existing retrieval systems are built around search engines and lack a deep understanding of the data within service systems, making them incapable of adapting to the professionalism and experience inherent in these systems.

[0078]

[0080] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present disclosure, and will not be described in detail here.

[0079]

[0081] FIG5 shows an overall flow chart of a document retrieval method provided by an embodiment of the present disclosure. Referring to FIG5 , the method may include the following steps.

[0080]

[0082] Step 110: The user account enters a search text, and sends a document search request based on the entered search text.

[0081]

[0083] Step 120: In response to the document retrieval request, determine whether the retrieved text is natural language that cannot be extracted. If the retrieved text is identified as specific data or categorized data from the internal system, execute steps 140-141 to perform feature extraction and data matching. If it is identified as natural language, execute steps 130-131 to invoke the large language model for processing.

[0082]

[0084] Step 130 - Step 131: Call the large language model to identify the user's intention information, and convert the user's intention information into system-recognizable feature data pairs (ie, search terms), so that they can be input into the information fusion retrieval interface.

[0083]

[0085] Step 140 - Step 141: The system uses word segmentation, string matching and other methods to extract type or service data to obtain feature data pairs (ie, search terms), which can then be input into the information fusion search interface.

[0084]

[0086] Step 150: The unified fusion query interface is responsible for processing the input feature data pairs, thereby performing multiple parallel query tasks. Among them, steps 160 to 162, steps 170 to 172, and steps 180 to 182 are parallel processes.

[0085]

[0087] Step 160: The feature data is directly input into the document retrieval system for query.

[0086]

[0088] Step 161: Query the search engine's inverted index to see whether there are document keywords that match the feature data pair.

[0087]

[0089] Step 162: Return the search results and the score of the query relevance.

[0088]

[0090] Step 170: Convert the input feature data pairs into features to obtain vectorized representations of the feature data pairs.

[0089]

[0091] Step 171: Input the feature vector obtained in step 170 into the vector retrieval system, and perform matching in the vector retrieval system to obtain the matched feature vector with the highest similarity.

[0090]

[0092] Step 172: Return the search result corresponding to the feature vector with the highest similarity and the query similarity score.

[0091]

[0093] Step 180: Convert the input feature data into interface data of the internal data query interface, and call the internal service interface for retrieval.

[0092]

[0094] Step 181: Integrate the data returned by the service interface.

[0093]

[0095] Step 182: Return the integrated search results and keyword overlap.

[0094]

[0096] Step 190: Comprehensively evaluate and sort the search results returned by the above parallel tasks according to their respective scores, and finally return the integrated results.

[0095]

[0097] Please refer to FIG6 , which shows a schematic diagram of the structure of a document retrieval device provided by an embodiment of the present disclosure. The device can be implemented by software, hardware, or a combination of both, and can become all or part of an electronic device. The device includes:

[0096]

[0098] Processing module 601 is configured to respond to a document search request including target search text sent by a user account, process the target search text, and obtain a search term in a preset format, where the preset format is a data format recognizable by a preset search interface, which is an interface capable of performing searches using multiple search paths;

[0097]

[0099] A retrieval module 602 is configured to perform retrieval using multiple retrieval paths based on a retrieval term and a preset retrieval interface, and obtain document retrieval results corresponding to each retrieval path;

[0098]

[0100] The sending module 603 is used to send the document search results corresponding to the multiple search paths to the user account.

[0099]

[0101] In another embodiment of the present disclosure, the processing module 601 is configured to identify text content attributes of a target search text in response to a document search request; and process the target search text in different processing modes according to the text content attributes of the target search text to obtain search terms.

[0100]

[0102] In another embodiment of the present disclosure, the processing module 601 is configured to perform word segmentation and string matching on the target search text to obtain search terms when the text content attribute of the target search text is specific data or category data within the system.

[0101]

[0103] In another embodiment of the present disclosure, the processing module 601 is configured to, when the text content attribute of the target search text is natural language, call a large language model to recognize the target search text and obtain a search term.

[0102]

[0104] In another embodiment of the present disclosure, processing module 601 is used to call a large language model to identify the target search text and obtain multiple candidate intent information; in response to a user account's selection operation for any candidate intent information, obtain the candidate intent information selected by the user account; and process the candidate intent information to obtain a search term.

[0103]

[0105] In another embodiment of the present disclosure, the retrieval module 602 is configured to process search terms based on a preset search interface to obtain search path characteristic data corresponding to different search paths, where the search path characteristic data is required for performing a search using the corresponding search path; and perform a search based on the search path characteristic data corresponding to the different search paths to obtain document search results corresponding to each search path.

[0106] In another embodiment of the present disclosure, the retrieval module 602 is configured to use a search term directly as retrieval path feature data corresponding to a document retrieval path based on a preset retrieval interface; based on the search term, query a document database for document keywords matching the search term, where the document database is configured to store the correspondence between document keywords and documents using an inverted document index; and when a document keyword matching the search term is stored in the document database, use the document corresponding to the matching document keyword as a document retrieval result corresponding to the document retrieval path.

[0104]

[0107] In another embodiment of the present disclosure, the retrieval module 602 is configured to obtain a target feature vector corresponding to a search term based on a preset retrieval interface, and use the target feature vector as retrieval path feature data corresponding to a vector retrieval path. Based on the target feature vector, the retrieval module 602 searches a vector database for a feature vector whose similarity to the target feature vector exceeds a preset threshold, wherein the vector database is configured to store correspondences between feature vectors and documents. When the vector database stores a feature vector whose similarity to the target feature vector exceeds a preset threshold, the document corresponding to the retrieved feature vector is used as a document retrieval result corresponding to the vector retrieval path.

[0105]

[0108] In another embodiment of the present disclosure, the retrieval module 602 is configured to convert a search term into interface data recognizable within the system based on a preset retrieval interface, and use the interface data as retrieval path feature data corresponding to an internal interface retrieval path; based on the interface data, call a corresponding service interface within the system to perform a retrieval, and obtain document retrieval results corresponding to the internal interface retrieval path.

[0106]

[0109] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0107]

[0110] FIG7 shows a structural block diagram of an electronic device 700 provided by an exemplary embodiment of the present disclosure. Generally, the electronic device 700 includes: a processor 701 and a memory 702.

[0108]

[0111] Processor 701 can be implemented in hardware using at least one of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), and a PLA (Programmable Logic Array). Processor 701 may also include a main processor and a coprocessor. The main processor is used to process data in an awake state; the coprocessor is a low-power processor used to process data in a standby state. In some embodiments, processor 701 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing content required to be displayed on the display screen. In some embodiments, processor 701 may also include an artificial intelligence processor for handling computational operations related to machine learning.

[0109]

[0112] The memory 702 may include one or more computer-readable storage media. The computer-readable storage media may be non-transitory computer-readable storage media. For example, the non-transitory computer-readable storage media may be a CD-ROM (Compact Disc Read-Only Memory), a ROM, a RAM (Random Access Memory), a magnetic tape, a floppy disk, an optical data storage device, etc. The computer-readable storage media may store at least one computer program. When the at least one computer program is executed, the document retrieval method may be implemented.

[0110]

[0113] Of course, the electronic device described above may also include other components, such as input / output interfaces and communication components. The input / output interface provides an interface between the processor and a peripheral interface module, which may be an output device or an input device. The communication component is configured to facilitate wired or wireless communication between the electronic device and other devices.

[0111]

[0114] Those skilled in the art will appreciate that the structure shown in FIG. 7 does not limit the electronic device 700 and may include more or fewer components than shown, or combine certain components, or adopt a different component arrangement.

[0112]

[0115] An embodiment of the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium stores at least one computer program, and when the at least one computer program is executed by a processor, the document retrieval method described above can be implemented.

[0113]

[0116] An embodiment of the present disclosure provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it can implement the above-mentioned document retrieval method.

[0114]

[0117] The above embodiments are merely intended to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they may modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein with equivalents. However, such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present disclosure.

Claims

Claims 1. A document retrieval method, comprising: In response to a document search request including target search text sent by a user account, processing the target search text to obtain a search term in a preset format, where the preset format is a data format recognizable by a preset search interface, and the preset search interface is an interface capable of performing searches using multiple search paths; performing searches using the multiple search paths based on the search term and the preset search interface, and obtaining document search results corresponding to each search path; The document retrieval results corresponding to the multiple retrieval paths are sent to the user account.

2. The method according to claim 1, wherein: The responding to a document retrieval request containing a target retrieval text sent by a user account, processing the target retrieval text to obtain a retrieval term in a preset format includes: identifying text content attributes of the target retrieval text in response to the document retrieval request; and processing the target retrieval text in different processing modes according to the text content attributes of the target retrieval text to obtain the retrieval term.

3. The method according to claim 2, wherein: The target search text is processed in different ways according to the text content attributes of the target search text to obtain the search term, including: when the text content attributes of the target search text are specific data or category data within the system, the target search text is subjected to word segmentation processing and string matching processing to obtain the search term.

4. The method according to claim 2, wherein: The step of processing the target search text in different ways according to the text content attributes of the target search text to obtain the search term includes: when the text content attribute of the target search text is natural language, calling a large language model to recognize the target search text to obtain the search term.

5. The method according to claim 4, wherein: The calling of the large language model to identify the target search text and obtain the search term includes: calling the large language model to identify the target search text and obtain multiple candidate intent information; in response to the user account's selection operation on any candidate intent information, obtaining the candidate intent information selected by the user account; processing the candidate intent information to obtain the search term.

6. The method according to claim 1, wherein: The step of performing a search using multiple search paths based on the search term and the preset search interface to obtain document search results corresponding to each search path includes: processing the search term based on the preset search interface to obtain search path characteristic data corresponding to different search paths, the search path characteristic data being data required for performing a search using the corresponding search path; Search is performed based on the search path feature data corresponding to different search paths to obtain document search results corresponding to each search path.

7. The method according to claim 6, wherein: The processing of the search terms based on the preset search interface to obtain search path characteristic data corresponding to different search paths includes: directly using the search terms as the search path characteristic data corresponding to the document search path based on the preset search interface; performing searches based on the search path characteristic data corresponding to different search paths to obtain document retrieval results corresponding to each search path, including: querying a document database for document keywords matching the search terms based on the search terms, the document database being used to store the correspondence between document keywords and documents in an inverted document manner; and when a document keyword matching the search term is stored in the document database, using the document corresponding to the matching document keyword as the document retrieval result corresponding to the document search path.

8. The method according to claim 6, wherein: The processing of the search terms based on the preset search interface to obtain search path feature data corresponding to different search paths includes: obtaining a target feature vector corresponding to the search term based on the preset search interface, and using the target feature vector as the search path feature data corresponding to the vector search path; performing retrieval based on the search path feature data corresponding to different search paths to obtain document retrieval results corresponding to each search path, including: querying a feature vector whose similarity with the target feature vector is higher than a preset threshold from a vector database based on the target feature vector, the vector database being used to store the correspondence between feature vectors and documents; and when a feature vector whose similarity with the target feature vector is higher than a preset threshold is stored in the vector database, using the document corresponding to the queried feature vector as the document retrieval result corresponding to the vector search path.

9. The method according to claim 6, wherein: The processing of the search terms based on the preset search interface to obtain search path characteristic data corresponding to different search paths includes: converting the search terms into interface data recognizable within the system based on the preset search interface, and using the interface data as search path characteristic data corresponding to the internal interface search path; performing searches based on the search path characteristic data corresponding to different search paths to obtain document retrieval results corresponding to each search path includes: calling a corresponding service interface within the system based on the interface data to perform searches to obtain document retrieval results corresponding to the internal interface search path.

10. A document retrieval device, comprising: A processing module is configured to respond to a document retrieval request including target retrieval text sent by a user account, process the target retrieval text, and obtain a retrieval term in a preset format, where the preset format is a data format recognizable by a preset retrieval interface, and the preset retrieval interface is an interface capable of performing retrieval using multiple retrieval paths; a retrieval module is configured to perform retrieval using multiple retrieval paths based on the retrieval term and the preset retrieval interface, and obtain document retrieval results corresponding to each retrieval path; and a sending module is configured to send the document retrieval results corresponding to the multiple retrieval paths to the user account.

11. An electronic device comprising a processor and a memory; wherein: The memory stores at least one program code; the at least one program code is used to be called and executed by the processor to implement the document retrieval method according to any one of claims 1 to 9.

12. A computer-readable storage medium, wherein: The computer-readable storage medium stores at least one computer program, and when the at least one computer program is executed by a processor, it can implement the document retrieval method according to any one of claims 1 to 9.

13. A computer program product, comprising a computer program, wherein: When the computer program is executed by a processor, the document retrieval method according to any one of claims 1 to 9 can be implemented. 17

Images