Information search method and device based on large model, electronic equipment and storage medium

Abstract

The embodiment of the invention provides an information search method and device based on a large model, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence and information search. The method comprises the following steps: acquiring target sample data; performing model training on a preset basic information search model based on the target sample data to obtain an original information search model; performing knowledge graph construction based on the target sample data to obtain a target information knowledge graph; performing model adjustment on the original information search model based on the target information knowledge graph to obtain an initial information search model; performing model compression on the initial information search model to obtain a target information search model; performing keyword expansion on a pre-acquired original search keyword to obtain a target search keyword; and performing information search on the target search keyword through the target information search model to obtain target search information. According to the embodiment of the invention, the accuracy and timeliness of information search can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence and information search technology, and in particular to an information search method and device, electronic device and storage medium based on a large model. Background Art

[0002] Information search is an artificial intelligence technology that can find valuable information of interest to users from massive amounts of information. It can be applied to a variety of scenarios, such as information search in the education field, news report search, scientific research search, etc.

[0003] At present, in information search scenarios, pre-trained large language models are mainly used as information search models for information search. However, in actual use scenarios, due to the wide range of information sources and diverse expressions, it is difficult for information search models to retrieve accurate and required information.

[0004] Therefore, how to improve the accuracy of information search has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of the embodiments of the present application is to propose an information search method and device, an electronic device and a storage medium based on a large model, aiming to improve the accuracy of information search.

[0006] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes an information search method based on a large model, the method comprising:

[0007] Obtain target sample data;

[0008] Performing model training on a preset basic information search model based on the target sample data to obtain an original information search model;

[0009] Construct a knowledge graph based on the target sample data to obtain a target information knowledge graph;

[0010] Based on the target information knowledge graph, the original information search model is adjusted to obtain an initial information search model;

[0011] Performing model compression on the initial information search model to obtain a target information search model;

[0012] Perform keyword expansion on the original search keywords obtained in advance to obtain target search keywords;

[0013] The target information search model is used to perform information search on the target search keyword to obtain target search information.

[0014] In some embodiments, constructing a knowledge graph based on the target sample data to obtain a target information knowledge graph includes:

[0015] Performing entity extraction on the target sample data to obtain original entity information;

[0016] Extracting entity relationships from the target sample data to obtain original entity relationship information; wherein the original entity relationship information is used to characterize the relationship of the original entity information;

[0017] Constructing a graph based on the original entity relationship information and the original entity information to obtain an initial information knowledge graph;

[0018] The initial information knowledge graph is indexed and constructed to obtain the target information knowledge graph.

[0019] In some embodiments, compressing the initial information search model to obtain a target information search model includes:

[0020] Pruning the initial information search model to obtain a pruned information search model;

[0021] Performing distillation processing on the pruned information search model to obtain a distilled information search model;

[0022] The distilled information search model is quantized to obtain the target information search model.

[0023] In some embodiments, the step of performing keyword expansion on the pre-acquired original search keyword to obtain the target search keyword includes:

[0024] Obtaining user historical search records, and performing context analysis based on the user historical search records to obtain historical search information;

[0025] Perform synonymous concept query based on the original search keyword to obtain synonymous keywords;

[0026] Keyword rewriting is performed based on the original search keyword, the historical search information and the synonymous keyword to obtain the target search keyword.

[0027] In some embodiments, performing information search on the target search keyword through the target information search model to obtain target search information includes:

[0028] Perform information query based on the target information search model and the target search keyword to obtain original search information;

[0029] The original search information is structured to obtain the target search information.

[0030] In some embodiments, the original search information includes a plurality of original search sub-information; and the structural processing of the original search information to obtain the target search information includes:

[0031] Performing deduplication processing on each of the original search sub-information to obtain initial search sub-information;

[0032] Based on the target search keyword, a correlation calculation is performed on each of the initial search sub-information to obtain a search correlation score sequence;

[0033] Calculate the source score of each of the initial search sub-information based on the preset data source weight to obtain a search source score sequence;

[0034] Calculating the timeliness score of each of the initial search sub-information based on the preset timeliness weight to obtain a search timeliness score sequence;

[0035] Performing weighted calculation on the search relevance score sequence, the search source score sequence and the search timeliness score sequence to obtain a target search score sequence;

[0036] Performing content screening on the initial search sub-information based on the target search score sequence to obtain candidate search information;

[0037] The candidate search information is integrated to obtain the target search information.

[0038] In some embodiments, obtaining target sample data includes:

[0039] Get original information;

[0040] Performing data cleaning on the original information to obtain initial information;

[0041] Performing noise reduction processing on the initial information to obtain noise reduction information;

[0042] Performing format conversion on the noise reduction information to obtain target information;

[0043] The target information is labeled to obtain the target sample data.

[0044] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes an information search device based on a large model, the device comprising:

[0045] A sample data acquisition module is used to acquire target sample data;

[0046] A model training module, used to perform model training on a preset basic information search model based on the target sample data to obtain an original information search model;

[0047] A knowledge graph construction module is used to construct a knowledge graph based on the target sample data to obtain a target information knowledge graph;

[0048] A model adjustment module, used to adjust the original information search model based on the target information knowledge graph to obtain an initial information search model;

[0049] A model compression module, used for compressing the initial information search model to obtain a target information search model;

[0050] A search keyword expansion module is used to expand the original search keywords obtained in advance to obtain target search keywords;

[0051] The information search module is used to perform information search on the target search keyword through the target information search model to obtain target search information.

[0052] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0053] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0054] The information search method and device based on the large model, the electronic device and the storage medium proposed in the present application obtain the target sample data, and construct the knowledge graph based on the target sample data to obtain the target information knowledge graph; the preset basic information search model is trained based on the target sample data to obtain the original information search model, which can enhance the professionalism of the model in the field of information search; the original information search model is adjusted based on the target information knowledge graph to obtain the initial information search model, which can enrich the semantic understanding ability of the model; the initial information search model is compressed to obtain the target information search model, which helps to ensure the efficient operation of the model and reduce the consumption of computing resources. Furthermore, the original search keywords obtained in advance are expanded to obtain the target search keywords, which broadens the search scope and makes the search content richer; finally, the target search keywords are searched for information through the target information search model to obtain the target search information, which improves the accuracy and timeliness of the information search. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1is a flow chart of a large model-based information search method provided in an embodiment of the present application;

[0056] Figure 2 yes Figure 1 Flow chart of step S101 in FIG.

[0057] Figure 3 yes Figure 1 Flow chart of step S103 in FIG.

[0058] Figure 4 yes Figure 1 Flow chart of step S105 in FIG.

[0059] Figure 5 yes Figure 1 Flow chart of step S106 in FIG.

[0060] Figure 6 yes Figure 1 Flow chart of step S107 in FIG.

[0061] Figure 7 yes Figure 6 Flowchart of step S602 in FIG.

[0062] Figure 8 It is a structural schematic diagram of an information search device based on a large model provided in an embodiment of the present application;

[0063] Fig. 9 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0065] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0066] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0067] First, some nouns involved in this application are analyzed:

[0068] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0069] Natural language processing (NLP): NLP uses computers to process, understand and apply human languages ​​(such as Chinese, English, etc.). NLP is a branch of artificial intelligence and an interdisciplinary subject between computer science and linguistics. It is often referred to as computational linguistics. Natural language processing includes grammatical analysis, semantic analysis, and text understanding. Natural language processing is often used in technical fields such as machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, public opinion analysis and opinion mining. It involves data mining related to language processing, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computing.

[0070] Information Extraction: A text processing technology that extracts specified types of entity, relationship, event and other factual information from natural language text and forms structured data output. Information extraction is a technology that extracts specific information from text data. Text data is composed of some specific units, such as sentences, paragraphs, and chapters. Text information is composed of some small specific units, such as characters, words, phrases, sentences, paragraphs, or a combination of these specific units. Extracting noun phrases, names, place names, etc. from text data are all text information extraction. Of course, the information extracted by text information extraction technology can be various types of information.

[0071] Knowledge Graph is a graph-based data structure that displays entities, concepts and their relationships in the form of a graph, so that the relevance and hierarchy of information can be clearly displayed. In a knowledge graph, nodes usually represent entities or concepts, such as people, places, events, etc., while edges represent the associations or attributes between these entities or concepts. By constructing a knowledge graph, knowledge can be understood and applied more intuitively, and knowledge can be effectively organized and used efficiently. Knowledge graphs are widely used in search engines, intelligent question and answer, recommendation systems and other fields, providing strong support for the development of artificial intelligence. They can not only improve the efficiency of information retrieval, but also deepen the machine's ability to understand human language and promote the continuous advancement of intelligent technology.

[0072] Large Language Model (LLM) is an artificial intelligence model based on deep learning and natural language processing technology. Large language models are trained with massive amounts of text data to learn common patterns and structures of language, thus having powerful natural language understanding and generation capabilities. Large language models are able to handle complex natural language tasks such as text generation, question-answering systems, machine translation, etc., and perform well in these tasks. Large language models are characterized by their large scale, usually with billions or more parameters, which enables large language models to capture rich features and patterns of language. At the same time, large language models also face challenges such as high computing resource requirements, limited semantic understanding, and possible ethical and bias issues.

[0073] Model compression is a technical means in the field of machine learning that aims to reduce the size, complexity and computational complexity of machine learning models. Specifically, model compression uses a series of optimization methods, such as parameter pruning, model distillation, model sparsification, weight sharing and low-precision calculations, to streamline the model structure, reduce the number of model parameters or reduce the number of model storage quantization bits, thereby significantly reducing the model's storage requirements and computational complexity while keeping the model performance basically unchanged. Model compression technology enables large models to achieve efficient and accurate reasoning and application on resource-constrained devices (such as mobile devices, edge devices, etc.), promoting the popularization and application of machine learning technology in more scenarios.

[0074] Information search is an artificial intelligence technology that can find valuable information of interest to users from massive amounts of information. It can be applied to a variety of scenarios, such as information search in the education field, news report search, scientific research search, etc.

[0075] At present, in information search scenarios, pre-trained large language models are mainly used as information search models for information search. However, in actual use scenarios, due to the wide range of information sources and diverse expressions, it is difficult for information search models to fully adapt to the search needs in the field of information search. The quality of search results is uneven, and it is difficult to retrieve accurate and satisfactory information.

[0076] Based on this, the embodiments of the present application provide a large model-based information search method and device, electronic device and storage medium, aiming to improve the accuracy of information search.

[0077] The information search method and device based on a large model, electronic device and storage medium provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the information search method based on a large model in the embodiments of the present application is described.

[0078] The information search method based on a large model provided in the embodiment of the present application relates to the field of artificial intelligence and information search technology. The information search method based on a large model provided in the embodiment of the present application can be applied to a terminal, can also be applied to a server side, and can also be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or it can be configured as a server cluster or a distributed system composed of multiple physical servers, and can also be configured as 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, CDN, and big data and artificial intelligence platforms; the software can be an application that implements an information search method based on a large model, etc., but is not limited to the above forms.

[0079] The present application can be used in many general or special 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. The present 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. The present application can also be practiced in distributed computing environments, in which 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.

[0080] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0081] Figure 1 is an optional flow chart of the information search method based on a large model provided in an embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S107.

[0082] Step S101, obtaining target sample data;

[0083] Step S102, training a preset basic information search model based on the target sample data to obtain an original information search model;

[0084] Step S103, constructing a knowledge graph based on the target sample data to obtain a target information knowledge graph;

[0085] Step S104, adjusting the original information search model based on the target information knowledge graph to obtain an initial information search model;

[0086] Step S105, compressing the initial information search model to obtain a target information search model;

[0087] Step S106, performing keyword expansion on the original search keyword obtained in advance to obtain a target search keyword;

[0088] Step S107: performing information search on the target search keyword through the target information search model to obtain target search information.

[0089] Steps S101 to S107 shown in the embodiment of the present application are to obtain target information knowledge graph by acquiring target sample data and constructing knowledge graph based on target sample data; to obtain original information search model by model training based on target sample data, which can enhance the professionalism of the model in the field of information search; to obtain initial information search model by model adjustment based on target information knowledge graph, which can enrich the semantic understanding ability of the model; to obtain target information search model by model compression on initial information search model, which helps to ensure the efficient operation of the model and reduce the consumption of computing resources. Furthermore, keyword expansion is performed on the original search keywords obtained in advance to obtain target search keywords, which broadens the search scope and makes the search content richer; finally, information search is performed on target search keywords through target information search model to obtain target search information, which improves the accuracy and timeliness of information search.

[0090] See also Figure 2 In some embodiments, step S101 may include but is not limited to steps S201 to S205:

[0091] Step S201, obtaining original information;

[0092] Step S202, performing data cleaning on the original information to obtain initial information;

[0093] Step S203, performing noise reduction processing on the initial information to obtain noise reduction information;

[0094] Step S204, converting the format of the noise reduction information to obtain target information;

[0095] Step S205: label the target information to obtain target sample data.

[0096] Steps S201 to S205 shown in the embodiment of the present application improve the quality of target sample data and reduce noise by performing data cleaning, noise reduction, format conversion and labeling on the original information, thereby providing a solid data foundation for information search and model training and helping to improve the accuracy of information search.

[0097] In step S201 of some embodiments, the original information is public information collected in advance from multiple source channels. For example, in a financial scenario, the original information may be financial news. In other scenarios, the original information may also be news, etc., where the source channels include but are not limited to: publications, academic papers, forums, news media, and verified online resources; public information includes but is not limited to: public reports, entertainment news, economic news, military news, etc., which can cover the information of relevant scenarios as comprehensively as possible, thereby helping to improve the accuracy of model search.

[0098] In addition, the original information may be text information, image information, audio information, video information or other types of data to provide more comprehensive information support.

[0099] In one embodiment, the raw information is on the order of hundreds of billions.

[0100] In step S202 of some embodiments, data cleaning is performed on the original information, such as processing special characters, punctuation marks, unified terms and abbreviations, etc., which needs to be set in combination with the actual application scenario, but is not limited thereto.

[0101] In step S203 of some embodiments, the initial information is subjected to noise reduction processing to remove irrelevant messages and noise data, such as advertisements, copyright notices, etc., to obtain noise reduction information, which helps to improve the validity and availability of the data.

[0102] In step S204 of some embodiments, standardized text encoding (such as UTF-8 encoding) can be used to convert information of different formats (such as PDF, Word) in the noise reduction information into the same format (such as txt, JSON format), etc., so as to facilitate the model to read and learn relevant information.

[0103] In some embodiments, step S205 may include but is not limited to the following steps:

[0104] Perform paragraph / sentence segmentation on the target information to obtain segmentation information;

[0105] The segmentation information is labeled and processed to obtain the target sample data.

[0106] Specifically, the annotation tags may include, but are not limited to: news, editorial, article, analysis, summary, historical review, etc., which helps the model learn different types of relevant information.

[0107] In step S102 of some embodiments, the preset basic information search model is a pre-trained large language model that can perform basic information search tasks. However, the basic information search model is not trained for information in a specific domain scenario, and the search performance of the model may be poor. Therefore, it is necessary to train the basic information search model based on target sample data in a specific domain to obtain an original information search model with excellent performance.

[0108] Among them, specific fields refer to sub-fields, including but not limited to finance, entertainment, Internet, education, etc., which need to be set up in combination with actual application scenarios.

[0109] In one embodiment, the basic information search model can be a GPT model or a Qianwen language model (such as the Qwen2.5 open source model). The model selection needs to be combined with the actual application scenario, but is not limited thereto. It should be noted that the pre-trained model usually selects a model size that matches the task scale.

[0110] Specifically, the model training process includes the following:

[0111] First, the target sample data is divided into training set, validation set and test set to ensure that each subset is representative and the annotation labels are balanced to avoid model overfitting.

[0112] During model training, model performance can be optimized by adjusting hyperparameters such as learning rate, batch size, number of training rounds, etc.

[0113] Furthermore, in order to ensure that the model can accurately capture the unique patterns of relevant information in a specific field, some specific task forms can be introduced during model training, such as generating summaries, answering questions, and classifying documents, to optimize the performance of the model in the field of information search. The specific implementation includes:

[0114] Generate summaries using sequence-to-sequence architectures or Transformer-based models.

[0115] Use the reading comprehension task format to develop a question-answering system to answer questions;

[0116] Use the text classification model to classify the target sample data and implement document classification tasks;

[0117] Throughout the process, by adjusting hyperparameters (such as learning rate, batch size, number of training rounds, etc.), selecting appropriate loss functions (mean square error loss function, cross entropy loss function, etc.) and evaluation indicators (accuracy, recall rate, etc.), and conducting sufficient training and verification, we ensure that the model can accurately capture the unique patterns of specific fields, thereby improving its performance on different task types.

[0118] In addition, early stopping and learning rate decay strategies can be used to prevent overfitting, and the learning progress of the model can be monitored by the performance on the validation set. During the entire model training process, the performance of the model on the test set is regularly evaluated to ensure that the model not only performs well on the training data, but also maintains a high generalization ability on unseen data.

[0119] It is understandable that the model training process in the above steps can actually be understood as fine-tuning of the large language model by adjusting the model parameters to meet the needs of information search in specific fields.

[0120] See also Figure 3 In some embodiments, step S103 may include but is not limited to steps S301 to S304:

[0121] Step S301, extracting entities from target sample data to obtain original entity information;

[0122] Step S302, extracting entity relationships from the target sample data to obtain original entity relationship information; wherein the original entity relationship information is used to characterize the relationship between the original entity information;

[0123] Step S303, constructing a graph based on the original entity relationship information and the original entity information to obtain an initial information knowledge graph;

[0124] Step S304, indexing and constructing the initial information knowledge graph to obtain the target information knowledge graph.

[0125] Steps S301 to S304 shown in the embodiment of the present application extract original entity information and original entity relationship information from target sample data, construct a graph based on the original entity relationship information and the original entity information to obtain an initial information knowledge graph, and then index the initial information knowledge graph to obtain a target information knowledge graph, which not only improves the structure and relevance of information, but also improves query efficiency and knowledge reasoning ability through index construction, making information retrieval and application more efficient and accurate.

[0126] In step S301 of some embodiments, natural language processing technology (such as named entity recognition NER, relationship extraction algorithm) can be used to perform sequence annotation on the target sample data to identify entities in the text and obtain original entity information.

[0127] For example, the original entity information may include, but is not limited to: business units, products, actions (such as publicity actions, marketing actions, etc.), locations, times, personnel (such as corporate legal persons, managers, etc.), etc.

[0128] In step S302 of some embodiments, based on the identified original entity information and text context information, a deep learning model (such as an attention mechanism, a graph neural network, etc.) is used to further extract the relationship between entities to obtain the original entity relationship information.

[0129] For example, the original entity relationship information may include, but is not limited to: the relationship between business units, the relationship between products and business units, the relationship between actions and business units, the relationship between personnel and business units, the relationship between locations and actions, etc.

[0130] In step S303 of some embodiments, multiple original entity information are connected based on the original entity relationship information to form a graph to obtain an initial information knowledge graph.

[0131] In some embodiments, a suitable database can be selected to store the knowledge graph, such as a graph database (such as Neo4j) or a relational database (such as MySQL), which is conducive to quickly extracting and using the knowledge graph.

[0132] In step S304 of some embodiments, in the knowledge graph, the index can quickly find specific entities or relationships, thereby improving query efficiency. Specifically, an index mechanism can be constructed for the initial information knowledge graph using inverted index, B-tree / B+ tree index, graph database index, finite state machine (FST) index, etc., to obtain the target information knowledge graph.

[0133] It is understandable that the target information knowledge graph not only contains basic entity information, but also covers rich contextual relationships. During real-time queries, the information in the knowledge graph can be quickly integrated, significantly improving the accuracy and richness of the query content.

[0134] In step S104 of some embodiments, the target information knowledge graph is used to further adjust the original information search model after model training, which can ensure that the model not only has extensive natural language understanding capabilities, but also has an in-depth understanding of the unique concepts and contextual relationships in specific fields, which helps to improve the performance of the model on specific tasks, such as document classification, analysis, etc.

[0135] Specifically, during the model adjustment process, the model will first retrieve relevant entities and relationships from the knowledge graph, and then input this information as context into the original information search model to perform specific task forms, such as document classification, analysis, report generation, and other tasks. Through continuous iteration and optimization, it ensures that the system can adapt to different task requirements and provide high-quality output.

[0136] It can be understood that after the model training in step S102 and the model adjustment in step S104, the initial information search model can perform tasks including but not limited to generating summaries, answering questions, classifying documents, analysis, and report generation, and can meet various needs, thereby improving the applicability of the model.

[0137] In some embodiments, the knowledge graph can be updated regularly to incorporate the latest relevant information to maintain the timeliness and practicality of information search.

[0138] See also Figure 4 In some embodiments, step S105 may include but is not limited to steps S401 to S403:

[0139] Step S401, pruning the initial information search model to obtain a pruned information search model;

[0140] Step S402, performing distillation processing on the pruned information search model to obtain a distilled information search model;

[0141] Step S403, quantizing the distilled information search model to obtain a target information search model.

[0142] Steps S401 to S403 shown in the embodiment of the present application can effectively reduce the complexity of the model and the consumption of computing resources by pruning, distilling and quantizing the initial information search model, while maintaining the key search capabilities of the model, thereby obtaining a more efficient and lightweight target information search model, which is conducive to improving the real-time and accuracy of information search.

[0143] Moreover, the order of pruning, distillation, and quantization helps ensure that each stage can effectively serve the desired goal, that is, to achieve efficient deployment of the model while maintaining the highest possible accuracy.

[0144] In step S401 of some embodiments, some neuron connections in the initial information search model are pruned to reduce the size of the model, thereby obtaining a pruned information search model;

[0145] Specifically, the pruning process includes but is not limited to the following steps:

[0146] The importance of neurons in the initial information search model is scored to obtain the neuron importance score;

[0147] Based on the preset importance threshold or pruning ratio, combined with the neuron importance score, determine which neurons need to be pruned and obtain candidate pruning neurons;

[0148] The initial information search model is pruned based on the candidate pruning neurons to obtain a pruned information search model.

[0149] After pruning, the pruned information search model needs to be fine-tuned and evaluated to ensure that the pruned model has the capabilities and accuracy of the initial information search model.

[0150] In step S402 of some embodiments, the pruned information search model is used as a teacher model, and a student model is set to learn the performance characteristics of the teacher model while having an output format and dimension similar to the teacher model, thereby simplifying the model.

[0151] In step S403 of some embodiments, the distilled information search model is quantized, and the floating-point weights in the model are converted into low-precision integers or fixed-point numbers to obtain a target information search model, which can significantly reduce the storage space and computational overhead of the target information search model.

[0152] In some embodiments, the distillation information search model can be quantized using 8-bit quantization technology (INT8), or 4-bit quantization technology (INT4). The specific selection needs to be made in combination with the actual application scenario, but is not limited to this.

[0153] After step S403 in some embodiments, the large model-based information search method also includes: using technologies such as TensorFlow Lite, ONNX Runtime, etc. to convert the target information search model into a specific deployment format, deploying it on a target device (such as an embedded system, mobile device or cloud server), and verifying the deployed target information search model to ensure the reliability and practicality of the target information search model in a real environment.

[0154] It should be noted that the three model compression technologies illustrated in the above steps S401 to S403 are pruning technology, distillation technology and quantization technology, which can be selectively applied in combination with actual scenarios. For example, only distillation technology can be applied, or pruning technology and quantization technology can be applied simultaneously, or distillation technology and quantization technology can be applied simultaneously, but are not limited to this.

[0155] See also Figure 5 In some embodiments, step S106 may also include but is not limited to steps S501 to S503:

[0156] Step S501, obtaining a user's historical search record, and performing context analysis based on the user's historical search record to obtain historical search information;

[0157] Step S502, performing a synonym concept query based on the original search keyword to obtain a synonym keyword;

[0158] Step S503: rewrite the keywords based on the original search keywords, historical search information and synonymous keywords to obtain target search keywords.

[0159] Steps S501 to S503 shown in the embodiment of the present application parse the historical search records to determine the user's historical search information, thereby referring to the content that the user may be interested in; then use synonymous concept queries to broaden the search scope and obtain synonymous keywords; finally, rewrite the original search keywords in combination with the historical search information and synonymous keywords, thereby generating target search keywords that are closer to the user's intentions and meet the characteristics of a specific field, which helps to improve the relevance and accuracy of information search.

[0160] It should be noted that the original search keywords are the keywords entered by the user on the target device (such as an embedded system, mobile device or cloud server) where the target information search model is deployed. They are usually relatively simple and common words, so they need to be expanded and rewritten to obtain target search keywords that meet the characteristics of specific fields.

[0161] In step S501 of some embodiments, the user's historical search records may be obtained, and context analysis may be performed based on the user's historical search records to obtain historical search information, thereby obtaining content that the user may be interested in. For example:

[0162] If the user's historical search records contain content related to a certain company, or the current query context is raised when discussing a company related to a certain industry, it can be assumed that the user may be interested in a specific industry, and the historical search information can be derived as "company related to a certain industry", "technology in a certain industry", etc.

[0163] In step S502 of some embodiments, a synonymous concept query may be performed based on the original search keyword to obtain synonymous keywords, thereby expanding the keyword range. For example:

[0164] If the original search keyword input by the user is "Internet", then related synonymous concepts may be "Internet", "Internet", "Internet technology" and the like.

[0165] In some embodiments, the original search keywords may also be rewritten based on specific domain terms, for example:

[0166] If the original search keyword is "modern trading strategy", the rewritten search keyword can be: "21st century trading strategy", "informationized trading strategy", "Internet trading strategy", etc. Because "modern trading strategy" is a relatively broad concept, by rewriting, more specific and relevant terms are introduced, which helps to narrow the search scope.

[0167] In step S503 of some embodiments, the original search keyword may be rewritten by combining historical search information, synonymous keywords, or one or more of specific domain terms to obtain a target search keyword.

[0168] For example:

[0169] Embodiment 1: rewriting keywords based on historical search information and specific domain terms to obtain target search keywords.

[0170] Original search keywords: Modern trading strategies;

[0171] Target search keywords: information technology transaction strategy, Internet transaction strategy, blockchain transaction strategy;

[0172] Explanation: If the user has browsed content related to the Internet before, or the current query context is raised in the context of discussing the development of the Internet, it can be assumed that the user may be interested in Internet-related content. Therefore, by introducing these more specific subtopics and rewriting them in conjunction with Internet domain terms, it is possible to provide results that are more in line with user expectations.

[0173] Embodiment 2: Rewrite keywords based on synonymous keywords and specific field terms to obtain target search keywords.

[0174] Original search keywords: Modern trading strategies;

[0175] Target search keywords: 21st century trading strategy, information-based trading strategy;

[0176] Explanation: "Modern trading strategy" is a relatively broad and inaccurate concept. Therefore, by searching for synonymous concepts such as "21st century trading strategy" and "information-based trading strategy", we can ensure the accuracy of the search keywords and conform to the expression habits in specific fields, while also providing results that are more in line with user expectations.

[0177] In some embodiments, keyword expansion may be implemented based on one or more of the following rules:

[0178] Domain-specific term expansion: identifying and replacing common vocabulary with domain-specific terms;

[0179] Introduction of synonymous concepts: Introducing synonymous concepts based on knowledge graphs or domain expert systems;

[0180] Context-aware adjustments: Adjust search terms based on the user's historical query records or the context of the current query;

[0181] Question type identification and conversion: Convert natural language questions into a form that is more suitable for search engines to understand.

[0182] Entity Recognition and Linking: Identify key entities in the query and link to relevant database entries or knowledge bases.

[0183] Specifically, keyword expansion can be converted into a formulaic description:

[0184] Target search keywords = F (original search keywords, historical search information, synonymous keywords, specific field terms);

[0185] Among them, F can be a keyword expansion function implemented by a target information search model, and can also be a keyword expansion function implemented by a pre-trained large language model, but is not limited thereto.

[0186] See also Figure 6 In some embodiments, step S107 includes but is not limited to steps S601 to S602:

[0187] Step S601, performing information query based on the target information search model and the target search keywords to obtain original search information;

[0188] Step S602: Structural processing is performed on the original search information to obtain target search information.

[0189] Steps S601 to S602 shown in the embodiment of the present application obtain original search information by performing information query in target sample data and target knowledge graph based on target search keywords through a target information search model, and can efficiently and accurately filter out content related to the target search keywords from massive information and perform structured processing on the content, thereby improving the readability and utilization of the information and improving the accuracy of information search.

[0190] Before step S601 in some embodiments, the information search method based on the large model further includes:

[0191] Determine the target sample data and target knowledge graph as the target search source.

[0192] In step S601 of some embodiments, an information query is performed in a target search source based on a target search keyword through a target information search model to obtain original search information; wherein the original search information includes a plurality of original search sub-information.

[0193] Therefore, it is necessary to filter out high-quality information that best suits the current query intent from multiple original search sub-information and generate target search information.

[0194] See also Figure 7 In some embodiments, step S602 may include but is not limited to steps S701 to S704:

[0195] Step S701, performing deduplication processing on each original search sub-information to obtain initial search sub-information;

[0196] Step S702, performing relevance calculation on each initial search sub-information based on the target search keyword to obtain a search relevance score sequence;

[0197] Step S703, calculating the source score of each initial search sub-information based on the preset data source weight to obtain a search source score sequence;

[0198] Step S704, calculating the timeliness score for each initial search sub-information based on the preset timeliness weight to obtain a search timeliness score sequence;

[0199] Step S705, performing weighted calculation on the search relevance score sequence, the search source score sequence and the search timeliness score sequence to obtain a target search score sequence;

[0200] Step S706, screening the initial search sub-information based on the target search score sequence to obtain candidate search information;

[0201] Step S707: integrating the candidate search information to obtain target search information.

[0202] Steps S701 to S707 shown in the embodiment of the present application can effectively improve the accuracy and relevance of the search content, avoid the interference of duplicate information, and ensure that the target search information finally obtained is both refined and in line with the user's search needs by deduplicating, calculating relevance, calculating source scores, calculating timeliness scores, screening contents and integrating information for each search sub-information.

[0203] In step S701 of some embodiments, the same content can be identified through hash fingerprint technology, and semantic similarity calculation can be used to find highly similar content, and only one identical / similar content is retained, thereby achieving deduplication processing, avoiding interference from duplicate information, and improving the accuracy and efficiency of information retrieval.

[0204] In step S702 of some embodiments, a statistical method or a machine learning algorithm, such as a term frequency-inverse document frequency model (TF-IDF), can be used to measure the degree of match between the target search keyword and the vocabulary in the initial search sub-information, and the relevance score can be calculated accordingly.

[0205] In step S703 of some embodiments, the preset data source weight data needs to be set according to the actual application scenario, for example:

[0206] Sources include, but are not limited to: publications, academic papers, forums, news media, and online resources;

[0207] Among them, the source weight of the publication is set to 0.8, and the corresponding search source score is 1*0.8;

[0208] The source weight of an academic paper can be set to 0.7, and the corresponding search source score is 1*0.7;

[0209] The source weight of news media can be set to 0.6, and the corresponding search source score is 1*0.6;

[0210] The source weight of the forum can be set to 0.4, and the corresponding search source score is 1*0.4;

[0211] The source weight of online resources can be set to 0.3, and the corresponding search source score is 1*0.3.

[0212] In step S704 of some embodiments, the preset timeliness weight needs to be set according to the actual application scenario. For example:

[0213] For information with data validity within six months, the validity weight is set to 0.8, and the corresponding search validity score is 1*0.8;

[0214] For information with a validity period of six months to one year, the validity weight is set to 0.7, and the corresponding search validity score is 1*0.7;

[0215] For information with data validity between one and three years, the validity weight is set to 0.6, and the corresponding search validity score is 1*0.6;

[0216] For information with data validity between three and eight years, the validity weight is set to 0.5, and the corresponding search validity score is 1*0.5;

[0217] For information with a data validity period of more than eight years, the validity weight is set to 0.4, and the corresponding search validity score is 1*0.4.

[0218] In step S705 of some embodiments, a search relevance score sequence, a search source score sequence, and a search timeliness score sequence are weighted based on preset weighted calculation weights to obtain a target search score sequence, wherein the preset weighted calculation weights need to be set according to the actual application scenario, for example:

[0219] The weighted calculation weight of the search relevance score sequence is 0.4.

[0220] The weighted calculation weight of the search source score sequence is 0.3.

[0221] The weighted calculation weight of the search time rating sequence is 0.3.

[0222] The calculation formula of the target search score sequence is as follows:

[0223] Target search score sequence = 0.4*search relevance score sequence + 0.3*search source score sequence + 0.3 search timeliness score sequence.

[0224] In step S706 of some embodiments, the top N initial search sub-information can be screened from the target search score sequence as candidate search information; wherein N can be 10, 20, 30, etc., but is not limited thereto.

[0225] In addition, the target search score sequence can also be screened based on a preset score threshold, thereby screening the initial search sub-information to obtain candidate search information; wherein the preset score threshold needs to be set according to the actual application scenario, for example, set to 0.8, 0.75, etc., but is not limited thereto.

[0226] In step S707 of some embodiments, automatic summarization and topic clustering technology can be used to integrate multiple candidate search information to generate concise and logically clear target search information, remove redundancy and present it in an orderly manner, and ensure that the final displayed information is both comprehensive and refined to meet the user's query needs.

[0227] Specifically, multiple candidate search information can be arranged according to the target search score and based on the principle of scores from large to small, so that the most relevant target information can be obtained efficiently and accurately.

[0228] In some embodiments, the target search information can be fed back on the target device (such as an embedded system, mobile device or cloud server) where the target information search model is deployed, and displayed on the connected display front-end page to achieve real-time interaction with the user.

[0229] See also Figure 8 The embodiment of the present application further provides an information search device based on a large model, which can implement the above-mentioned information search method based on a large model, and the device includes:

[0230] The sample data acquisition module 801 is used to acquire target sample data;

[0231] The model training module 802 is used to perform model training on a preset basic information search model based on target sample data to obtain an original information search model;

[0232] The knowledge graph construction module 803 is used to construct a knowledge graph based on the target sample data to obtain a target information knowledge graph;

[0233] A model adjustment module 804 is used to adjust the original information search model based on the target information knowledge graph to obtain an initial information search model;

[0234] A model compression module 805 is used to compress the initial information search model to obtain a target information search model;

[0235] A search keyword expansion module 806 is used to expand the original search keywords obtained in advance to obtain target search keywords;

[0236] The information search module 807 is used to perform information search on the target search keyword through the target information search model to obtain the target search information.

[0237] The specific implementation of the large model-based information search device is basically the same as the specific implementation of the large model-based information search method described above, and will not be repeated here.

[0238] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned large model-based information search method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0239] See also Fig. 9 , Fig. 9 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0240] The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0241] The memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the large model-based information search method of the embodiment of this application;

[0242] Input / output interface 903, used to implement information input and output;

[0243] Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0244] A bus 905 that transmits information between various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);

[0245] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .

[0246] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned large model-based information search method is implemented.

[0247] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0248] The information search method and device, electronic device and storage medium based on the large model provided in the embodiments of the present application obtain target sample data, and construct a knowledge graph based on the target sample data to obtain a target information knowledge graph; perform model training on a preset basic information search model based on the target sample data to obtain an original information search model, which can enhance the professionalism of the model in the field of information search; perform model adjustment on the original information search model based on the target information knowledge graph to obtain an initial information search model, which can enrich the semantic understanding ability of the model; perform model compression on the initial information search model to obtain a target information search model, which helps to ensure the efficient operation of the model and reduce the consumption of computing resources. Furthermore, keyword expansion is performed on the original search keywords obtained in advance to obtain target search keywords, which broadens the search scope and makes the search content richer; finally, information search is performed on the target search keywords through the target information search model to obtain target search information, which improves the accuracy and timeliness of the search.

[0249] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0250] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0251] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0252] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0253] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0254] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0255] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. The mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0256] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0257] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0258] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0259] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A large model-based information search method, characterized in that: The method comprises: Obtain target sample data; Performing model training on a preset basic information search model based on the target sample data to obtain an original information search model; Construct a knowledge graph based on the target sample data to obtain a target information knowledge graph; Based on the target information knowledge graph, the original information search model is adjusted to obtain an initial information search model; Performing model compression on the initial information search model to obtain a target information search model; Perform keyword expansion on the original search keywords obtained in advance to obtain target search keywords; The target information search model is used to perform information search on the target search keyword to obtain target search information.

2. The method according to claim 1, characterized in that The knowledge graph is constructed based on the target sample data to obtain a target information knowledge graph, including: Performing entity extraction on the target sample data to obtain original entity information; Extracting entity relationships from the target sample data to obtain original entity relationship information; wherein the original entity relationship information is used to characterize the relationship of the original entity information; Constructing a graph based on the original entity relationship information and the original entity information to obtain an initial information knowledge graph; The initial information knowledge graph is indexed and constructed to obtain the target information knowledge graph.

3. The method according to claim 1, characterized in that The step of compressing the initial information search model to obtain a target information search model includes: Pruning the initial information search model to obtain a pruned information search model; Performing distillation processing on the pruned information search model to obtain a distilled information search model; The distilled information search model is quantized to obtain the target information search model.

4. The method according to claim 1, characterized in that: The keyword expansion of the pre-acquired original search keyword to obtain the target search keyword includes: Obtaining user historical search records, and performing context analysis based on the user historical search records to obtain historical search information; Perform synonymous concept query based on the original search keyword to obtain synonymous keywords; Keyword rewriting is performed based on the original search keyword, the historical search information and the synonymous keyword to obtain the target search keyword.

5. The method according to claim 1, characterized in that The step of performing information search on the target search keyword through the target information search model to obtain target search information includes: Perform information query based on the target information search model and the target search keyword to obtain original search information; The original search information is structured to obtain the target search information.

6. The method according to claim 5, characterized in that The original search information includes a plurality of original search sub-information; the structural processing of the original search information to obtain the target search information includes: Performing deduplication processing on each of the original search sub-information to obtain initial search sub-information; Based on the target search keyword, a correlation calculation is performed on each of the initial search sub-information to obtain a search correlation score sequence; Calculate the source score of each of the initial search sub-information based on the preset data source weight to obtain a search source score sequence; Calculating the timeliness score of each of the initial search sub-information based on the preset timeliness weight to obtain a search timeliness score sequence; Performing weighted calculation on the search relevance score sequence, the search source score sequence and the search timeliness score sequence to obtain a target search score sequence; Performing content screening on the initial search sub-information based on the target search score sequence to obtain candidate search information; The candidate search information is integrated to obtain the target search information.

7. The method according to any one of claims 1 to 6, characterized in that: The obtaining of target sample data comprises: Get original information; Performing data cleaning on the original information to obtain initial information; Performing noise reduction processing on the initial information to obtain noise reduction information; Performing format conversion on the noise reduction information to obtain target information; The target information is labeled to obtain the target sample data.

8. An information search device based on a large model, characterized in that: The device comprises: A sample data acquisition module is used to acquire target sample data; A model training module, used to perform model training on a preset basic information search model based on the target sample data to obtain an original information search model; A knowledge graph construction module is used to construct a knowledge graph based on the target sample data to obtain a target information knowledge graph; A model adjustment module, used to adjust the original information search model based on the target information knowledge graph to obtain an initial information search model; A model compression module, used to compress the initial information search model to obtain a target information search model; A search keyword expansion module is used to expand the original search keywords obtained in advance to obtain target search keywords; The information search module is used to perform information search on the target search keyword through the target information search model to obtain target search information.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images