Text information search method and device, electronic equipment and storage medium

Abstract

The invention discloses a text information search method and device, electronic equipment and a storage medium, and relates to the technical field of text information search. According to the technical scheme, the input text of the search task executed by the user on the basis of the internet is obtained, the input text is subjected to association expansion to obtain the text set of the input text, and the text set comprises the multiple keywords and / or statements associated with the input text, so that the input text is more efficient. According to the method, the preset AIGC model can be trained through the text set and the demand search result of the user, when the AIGC model accurately outputs the demand search result corresponding to the input text, it is indicated that the AIGC model can intelligently summarize and analyze the webpage search result, the to-be-searched task of the user is executed according to the trained AIGC model, and the search efficiency of the user is improved. According to the technical scheme provided by the invention, the efficiency and the accuracy of text information search are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of text information search, and in particular to a text information search method, device, electronic device and storage medium. Background Art

[0002] With the rapid development of the Internet, users often need to face a large number of search results when browsing the web. How to quickly and accurately extract key information from these search results has become a key issue in improving the efficiency of user information retrieval. In the prior art, although some browser extension tools can provide search result summary functions, the generated summary information is relatively brief, resulting in certain limitations in the summary and deviation from user needs. Summary of the invention

[0003] In view of the above technical problems, the present invention provides a text information search method, device, electronic device and storage medium, which can intelligently summarize and analyze web page search results, thereby improving the efficiency and accuracy of text information search.

[0004] The embodiment of the present invention provides the following solution:

[0005] In a first aspect, an embodiment of the present invention provides a method for searching text information, the method comprising:

[0006] Obtaining the input text of the user performing a search task based on the Internet;

[0007] Performing association expansion on the input text to obtain a text set of the input text, wherein the text set includes a plurality of keywords and / or sentences associated with the input text;

[0008] The preset AIGC model is trained according to the text collection and the user's demand search results until the AIGC model accurately outputs the demand search results corresponding to the input text;

[0009] The user's task to be searched is executed according to the trained AIGC model to obtain the user's target search text for the task to be searched.

[0010] In an optional embodiment, the input text is associated and expanded to obtain a text set of the input text, including:

[0011] Enter the input text into different search engines to obtain page search results of each search engine;

[0012] Prioritize all page search results to obtain at least one target search result;

[0013] Perform word frequency statistics, part-of-speech tagging, and vectorization on the target search results in order to obtain word vectors of the keywords associated with the input text;

[0014] The set consisting of the word vectors of all keywords is determined as a text set.

[0015] In an optional embodiment, all page search results are prioritized to obtain at least one target search result, including:

[0016] Obtain the priority of each page search result according to the number of outbound links and the number of inbound links represented by each page search result, wherein the number of outbound links is the number of other pages linked to the search result page, and the number of inbound links is the number of link jumps from the search result page to the seed page associated with the input text;

[0017] The page search results with a priority greater than the level threshold are determined as target search results.

[0018] In an optional embodiment, obtaining the priority of each page search result according to the number of outbound links and the number of inbound links represented by each page search result includes:

[0019] According to the formula Get the priority P(U) of each page search result, where S(U) is the number of inbound links and L(U) is the number of outbound links.

[0020] In an optional embodiment, a preset AIGC model is trained according to the text set and the user's demand search results until the AIGC model accurately outputs the demand search results corresponding to the input text, including:

[0021] Map the text collection and the demand search results to the same vector space, and calculate the similarity between the text collection and the demand search results;

[0022] The AIGC model is trained according to the similarity, and the model loss function is adjusted based on the prediction results of the AIGC model;

[0023] When the difference between the prediction result of the AIGC model and the demand search result is less than a preset condition, it is determined that the AIGC model has completed training.

[0024] In an optional embodiment, before determining that the AIGC model has completed training, the method further includes:

[0025] The AIGC model is fine-tuned according to the cross entropy loss function and gradient descent algorithm to reduce the prediction error of the AIGC model.

[0026] In an optional embodiment, executing the user's to-be-searched task according to the trained AIGC model to obtain the user's target search text for the to-be-searched task includes:

[0027] Compile the trained AIGC model into a WebAssembly module and configure it as a search plug-in for the user's browser;

[0028] Enter the search text of the task to be searched into the search plug-in and run the search plug-in;

[0029] According to the output result of the search plug-in, the target search text of the user for the task to be searched is obtained.

[0030] In a second aspect, an embodiment of the present invention further provides a text information search device, the device comprising:

[0031] An acquisition module is used to acquire input text of a user performing a search task based on the Internet;

[0032] An acquisition module, used for performing association expansion on the input text to obtain a text set of the input text, wherein the text set includes a plurality of keywords and / or sentences associated with the input text;

[0033] A training module is used to train a preset AIGC model according to a text set and a user's demand search results until the AIGC model accurately outputs the demand search results corresponding to the input text;

[0034] The execution module is used to execute the user's task to be searched according to the trained AIGC model to obtain the user's target search text for the task to be searched.

[0035] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising a processor and a memory, wherein the memory is coupled to the processor and stores instructions, and when the instructions are executed by the processor, the electronic device executes the steps of any one of the methods in the first aspect.

[0036] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods in the first aspect.

[0037] Compared with the prior art, the text information search method, device, electronic device and storage medium of the present invention have the following advantages:

[0038] The technical solution of the present invention obtains the input text of the user performing the search task based on the Internet, and associates and expands the input text to obtain a text set of the input text. Since the text set includes multiple keywords and / or sentences associated with the input text, the preset AIGC model can be trained through the text set and the user's demand search results. When the AIGC model accurately outputs the demand search results corresponding to the input text, it means that the AIGC model has been able to intelligently summarize and analyze the web page search results, and the user's task to be searched is executed according to the trained AIGC model to obtain the user's target search text for the task to be searched. The technical solution of the present invention intelligently summarizes and analyzes the web page search results according to the personalized needs of different types of users, which can help different types of users quickly identify and extract key information in the search results, and improve the efficiency and accuracy of text information search. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A flowchart of a text information search method provided by an embodiment of the present invention;

[0041] Figure 2 An overall logic diagram of a text information search method provided by an embodiment of the present invention;

[0042] Figure 3 A schematic diagram of the structure of a text information search device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention belong to the scope of protection of the embodiments of the present invention.

[0044] At present, the processing of search results is mostly represented in the form of brief summaries, lacking intelligent and personalized analysis capabilities. Therefore, it is necessary to develop a browser extension method based on artificial intelligence to achieve intelligent summary and analysis of web page search results. The following embodiment of the present invention will specifically explain how to achieve the above purpose.

[0045] See also Figure 1 , Figure 1 This is a flowchart of a text information search method provided by an embodiment of the present invention. The search method can be applied to a computer or server device, or can be run based on a user's mobile terminal, and is not specifically limited here. The search method includes:

[0046] S11. Obtaining input text of a user performing a search task based on the Internet.

[0047] Specifically, when users need to perform a search task based on the Internet, they usually enter some keywords or phrases to describe the information they want to find. The input text can be the question that the user searches for on the Internet, and its expression can be a sentence, or a phrase or keyword. The input text can be obtained based on the historical information of the search task, for example, the text entered by the user in the historical search task is obtained through the history record of the browser, and it is used as the above-mentioned input text. After obtaining the input text of the user performing the search task based on the Internet, step S12 is entered.

[0048] S12: performing association expansion on the input text to obtain a text set of the input text, wherein the text set includes a plurality of keywords and / or sentences associated with the input text.

[0049] Specifically, the purpose of association expansion is to expand the input text in terms of association to obtain more related texts. The input text can be split into multiple keywords and key words, multiple keywords and synonyms of the keywords can be found, and the search results can be determined as a text set. For example, key words or phrases can be extracted from the input text. These words are the core of understanding the intention of the input text, and synonyms or related words related to the keywords can be found to expand the search scope; the context of the keywords in the text can be considered to understand their specific meaning and use; based on keywords and context, they can be expanded to related topics or fields; the intention of the user's input can be analyzed to infer other relevant information that the user may be interested in; the types of content that the user may need, such as tutorials, news, comments, academic articles, etc., can be determined, and the text set of the input text can be obtained based on the final determination results.

[0050] In practical applications, although the conventional expansion method can obtain a text set associated with the input text, the accuracy of its data is insufficient, resulting in deviations in subsequent model training. Based on this, in a specific implementation, step S12 includes sub-steps 12-1 to 12-4, including:

[0051] 12-1. Input the input text into different search engines respectively to obtain page search results of each search engine.

[0052] Different search engines have different characteristics, and the search algorithms they use are partially different. For example, Google, Bing, DuckDuckGo, Baidu and other search engines have their own search algorithms, and some search engines may provide personalized search results based on the user's geographic location or search history. Enter the same input text in the search box of each search engine and perform the search task; each search engine will return a different list of page results; record the entry of each search result, including title, URL and short description, and use it as the page search result.

[0053] To further ensure the accuracy of page search results, the embodiment of the present invention uses crawler technology to obtain a large number of search result pages from search engines, and ensures data diversity through random walk strategies. For example, automated Web crawler technology is used in combination with a distributed crawling framework, and the Scrapy crawler framework is used to crawl search result pages from search engines such as Google, Bing, and DuckDuckGo. During the crawling process, the random walk theory of probability theory is used to configure an adaptive crawling strategy to ensure that page search results are evenly and widely obtained from multiple search engines.

[0054] 12-2. Prioritize all page search results to obtain at least one target search result. The page search results can be analyzed, and the priority of each page search result can be calculated based on the comprehensive characteristics of its characterization data, for example, the priority is determined based on the number of links in each page search result. All priorities are sorted, and page search results greater than the level threshold are determined as target search results. All page search results can be screened through priority sorting, and subsequent analysis can be performed on page search results with greater importance.

[0055] Exemplarily, sub-step 12-2 includes:

[0056] The first step is to obtain the priority of each page search result based on the number of outbound links and inbound links represented by each page search result. The number of outbound links refers to the number of other pages linked to the search result page. Generally, the fewer the number of outbound links, the more important the search result page is. If a page only links to a few pages, this may indicate that its content is very concentrated. The number of inbound links refers to the number of link jumps from the search result page to the seed page associated with the input text. The seed page refers to the page related to the topic of the input text. As the starting point for the crawler to crawl the network, the distance from the starting point to the search result page is calculated based on the link structure between the pages. The closer the distance, the more important or relevant the page is. A mapping relationship can be established based on the relationship between the number of outbound links, the number of inbound links and the priority. The corresponding priority is determined in the mapping relationship based on the number of outbound links and the number of inbound links represented by each page search result.

[0057] Of course, we can also use the formula Get the priority P(U) of each page search result, where S(U) is the number of inbound links and L(U) is the number of outbound links. After getting the priority of each page search result, proceed to the next step.

[0058] In the second step, the page search results with a priority greater than the level threshold are determined as the target search results. The level threshold can be set based on the actual needs of the user or the experience of the technician, and is not specifically limited here, as long as the page search results that meet the needs can be screened out. At this point, the target search results have been obtained.

[0059] 12-3. Perform word frequency statistics, part-of-speech tagging and vectorization processing on the target search results in turn to obtain word vectors of the keywords associated with the input text.

[0060] Specifically, the TF-IDF model can be used to calculate the weights of keywords in the target search results, and then part-of-speech tagging and named entity recognition can be performed, and finally the text data can be converted into a vector representation for subsequent training of the AIGC model.

[0061] The first step of data preprocessing is to apply the TF-IDF model to calculate keyword weights. The TF-IDF model is based on the following formula:

[0062] TF-IDF(t,d)=TF(t,d)×IDF(t,d)

[0063] Among them, TF(t,d) is the word frequency of word t in the target search results, IDF(t,d) is the inverse document frequency of word t, and TF-IDF(t,d) is the weight of word t.

[0064] The part-of-speech tagging algorithm can be used to tag the keywords that have completed the weight calculation, such as the statistical hidden Markov model (HMM) to tag the target search results. The conditional random field model (CRF) is used for named entity recognition to extract entity information in the text. The following is the optimization objective function of the conditional random field model:

[0065]

[0066] Among them, L(θ) is the named entity of part-of-speech tagging, θ is the model parameter, D is the training data set, p(y i |x,θ) is the output sequence y given the input sequence x i The conditional probability of , λ(θ) is the regularization term. It can be understood that named entity recognition aims to identify entities with specific meanings from text and classify them into predefined categories. These entities can be names, places, organizations, dates, times, values, proper nouns, etc.

[0067] Finally, a natural language processing algorithm, such as Word2Vec or BERT, is used to vectorize the keywords associated with the input text. The Word2Vec model learns word vectors using the following formula:

[0068]

[0069] Among them, loss is the loss value calculated by the above loss function, v i is the vector representation of the central word, v j is the vector representation of the context word, similarity is the similarity function between the center word and the context word, σ is the sigmoid function, and context(i) is the context word set of the center word i. Through vectorization processing, text data can be converted into a numerical form that can be processed by the AIGC model, providing a data basis for subsequent AIGC model training and optimization.

[0070] 12-4. The set consisting of the word vectors of all keywords is determined as a text set. The text set can be named accordingly based on the input text so that the two form a corresponding relationship. In order to improve the efficiency of data collection, the embodiment of the present invention can also introduce parallel processing technology. Through the MapReduce model, the data collection task is distributed to multiple nodes for parallel execution, thereby speeding up the data collection speed.

[0071] So far, a text collection has been obtained based on the input text through association expansion.

[0072] S13. The preset AIGC model is trained according to the text set and the user's demand search results until the AIGC model accurately outputs the demand search results corresponding to the input text.

[0073] Specifically, the embodiment of the present invention adopts the AIGC model in model selection because the model is a pre-trained language model based on Transformers. The Transformers model performs well in processing sequence data and can capture long-distance dependencies in text. The Transformers model uses the Self-Attention Mechanism to capture long-distance dependencies in text, and its mathematical expression is as follows:

[0074]

[0075] The Attention (Q, K, V) mechanism can capture the dependencies between elements in the sequence, so that the model can better understand the contextual information of the language; Q is the query vector, so as to find other words or tags related to it in the sequence; K is the key vector, which is used to determine the relevance of each word or tag in the sequence to the current query; V is the value vector, which will be aggregated into the output according to its relevance to the query; d k is the dimension of the key vector.

[0076] The elements in the text collection can be marked based on the user's demand search results to form training samples and test samples of the AIGC model. The AIGC model is trained based on the training samples. When the number of training times or the prediction results reach the preset target, the accuracy of the AIGC model in processing the input text search results is tested based on the test samples. When the test results of the AIGC model are the same as the user's demand search results, it means that the AIGC model has achieved the training purpose.

[0077] Exemplarily, step S13 includes sub-steps S13-1 to S13-3, which are described in detail as follows:

[0078] S13-1. Map the text set and the requirement search results to the same vector space, and calculate the similarity between the text set and the requirement search results.

[0079] The vector space model can be combined to map the text set and the demand search results into space vectors and map them into the same vector space, which helps the AIGC model to understand the search results while also considering the user's query intent. The present invention converts the search results and user queries into their vector representations through the TF-IDF model, and then uses cosine similarity to calculate the similarity between them. The formula is as follows:

[0080]

[0081] Among them, similarity(q,d) is the similarity, q is the vector of the demand search results, d is the vector representation of the text set, |q| is the modulus of the vector of the demand search results, and |d| is the modulus of the vector in the text set.

[0082] S13-2. Train the AIGC model based on similarity and adjust the model loss function based on the prediction results of the AIGC model. Use the pre-processed data to perform supervised learning training on the AIGC model so that it can perform tasks such as search result classification and keyword extraction. This is the core of intelligent analysis. During the training process, the following loss function is used for model optimization:

[0083]

[0084] Among them, L is the loss value calculated by the above loss function, y i is the true label, is the probability predicted by the model. Through the above method, the reinforcement learning algorithm is introduced to simulate the interaction process between users and search results, optimize the decision-making strategy of the model in search result analysis, and thus improve the quality and efficiency of the analysis.

[0085] S13-3. When the difference between the prediction result of the AIGC model and the demand search result is less than the preset condition, it means that the AIGC model is able to implement accurate processing based on the user's needs, and it is determined that the AIGC model has completed training.

[0086] To further improve the accuracy of AIGC model processing, the embodiment of the present invention fine-tunes the AIGC model according to the cross entropy loss function and the gradient descent algorithm to reduce the prediction error of the AIGC model. The cross entropy loss function has been described above, and the update rule of the gradient descent algorithm is as follows:

[0087]

[0088] Among them, θ is the model parameter, which comes from the weight of the conditional random field model; α is the learning rate, which is a hyperparameter in the AIGC model training process; is the gradient of the loss function with respect to the parameters.

[0089] The prediction results of multiple models are integrated through ensemble learning to further improve the generalization ability and robustness of the AIGC model. The bagging or boosting method is used to combine multiple Transformers-based models to improve the generalization ability and robustness of the model. The generalization error of ensemble learning can be expressed as:

[0090]

[0091] Among them, L 集成 is the error of ensemble learning, L 模型i is the error of the i-th model, and N is the number of models. The error L based on ensemble learning can be 集成 The training status of the AIGC model is determined. If the error is less than the preset value, it means that its accuracy has met the usage requirements, and it is determined that the AIGC model has completed training.

[0092] At this point, the AIGC model has been trained using text collections and user demand search results.

[0093] S14. Execute the user's task to be searched according to the trained AIGC model to obtain the user's target search text for the task to be searched.

[0094] Specifically, the AIGC model can be configured as a plug-in for the search engine on the browser and started to run it. The text to be searched of the task to be searched is input into the search engine, which is convenient for users to use and obtain their desired target search text. It should be noted that the search based on the trained AIGC model can be configured based on the actual situation of the interaction between the AIGC model and the Internet. For example, the task text of the task to be searched can be input into the AIGC model, which is processed by the AIGC model and then input into the search engine to implement the search and obtain the target search text; of course, the task text can also be input into the search engine, and the search results of the search engine can be input into the AIGC model, which is further processed by the AIGC model to obtain the target search text.

[0095] Exemplarily, step S14 includes sub-steps S14-1 to S14-3, which are specifically described as follows:

[0096] S14-1. Compile the trained AIGC model into a WebAssembly module and configure it as a search plug-in for the user's browser. Configure the user's browser extension tool as a plug-in architecture, which not only supports dynamic loading and updating of the AIGC model, but also facilitates subsequent functional expansion and maintenance. Compile the AIGC model into a WebAssembly module, develop a browser extension program that can load and run the WebAssembly module, provide a user interface for the browser extension, allow users to perform search operations, and install the developed extension program into the user's browser to make it a search plug-in.

[0097] The browser plug-in architecture design can also adopt a modular design to separate the AIGC model, data preprocessing module, user interface module, etc., so as to facilitate independent development and updating. The data preprocessing module is used for the above-mentioned text collection processing to output a text collection. The event-driven model is used to ensure efficient communication and rapid response between modules. The design of the front-end user interface takes into account the user experience, ensuring that users can intuitively view the search result analysis and interact.

[0098] S14-2. Input the text to be searched of the task to be searched into the search plug-in, and run the search plug-in. The text to be searched can be entered in the plug-in interface. The provision of a user configuration interface allows users to adjust the analysis parameters according to personal preferences, increasing the flexibility and personalization of the tool. The user configuration interface provides a user-friendly configuration interface, allowing users to adjust analysis parameters, such as keyword weight, summary length, etc. The user's configuration information is saved using the front-end storage method so that it can be automatically loaded the next time it is used. The implementation of a personalized recommendation system dynamically adjusts the display method of analysis results based on the user's historical behavior and preferences to better meet user needs.

[0099] S14-3. According to the output results of the search plug-in, obtain the user's target search text for the task to be searched. The target search text can be displayed in segments through an information box to facilitate user reading. Of course, list views, card views, network views, etc. can also be displayed based on user needs. The display method of the list view includes title, summary and link; the card view is displayed in the form of cards, usually containing richer information, such as pictures, titles, descriptions and links; the network view displays the results in the form of a grid, and each grid contains a thumbnail and a brief description. Different display methods can better meet the user's text reading habits, thereby improving the user's reading efficiency of the target search text. It should be further explained that the technical solution included in the embodiment of the present invention can also play the target search text through a voice plug-in to interact with the user in a diverse and personalized way.

[0100] The following embodiment of the present invention will generally describe how to implement the search of text information. Figure 2 , Figure 2 This is the overall flow chart of the search method. Specifically, it includes:

[0101] S201, data collection: obtaining input text according to the user's historical records in the search task, and inputting the input text into different search engines to obtain page search results of each search engine.

[0102] S202, preprocessing, including word frequency statistics, part-of-speech tagging, name recognition and vectorization processing of page search results to obtain a text collection.

[0103] S203, AIGC model training. Model training is performed based on the text set until the AIGC model accurately outputs the required search results corresponding to the input text.

[0104] S204, model optimization. Specifically, it includes fine-tuning the AIGC model through the cross entropy loss function and the gradient descent algorithm, and implementing accuracy prediction of the AIGC model through ensemble learning.

[0105] S205: Model update: Update the AIGC model after training.

[0106] S206, browser extension development. Specifically includes configuration architecture plug-ins, user interface and function implementation settings.

[0107] S207: When implementing the function, it specifically includes real-time analysis, WebAssembly module compilation and custom settings. The custom settings include configuration interface and personalized settings.

[0108] The following example illustrates the overall implementation strategy of text information search. For example, a browser extension tool called "SmartSummarizer" is developed to help users quickly summarize and analyze web search results. If you want to know the relevant information about "new coronavirus vaccine", first enter "new coronavirus vaccine" as input text into different search engines. The crawler obtains a large number of web pages containing the keyword "new coronavirus vaccine" from search engines such as Google, Bing, and DuckDuckGo, including news, research papers, popular science articles, etc.

[0109] Then use the TF-IDF model to analyze the keyword weights. For example, words such as "new coronavirus", "vaccine", and "effectiveness" will be more important. Perform part-of-speech tagging and named entity recognition on the text to extract entity information such as "COVID-19", "mRNA vaccine", and "adenovirus vector vaccine". Convert the text into a vector representation to facilitate model processing.

[0110] Select a pre-trained language model based on Transformers. Use the pre-processed dataset to train the model so that it can understand the relevant knowledge of "new coronavirus vaccine" and perform tasks such as keyword extraction and summary generation. Use reinforcement learning to simulate the interaction process between users and search results, and optimize the model's decision-making strategy when analyzing "new coronavirus vaccine" search results, such as paying more attention to entity information that users may be interested in.

[0111] Users install the "SmartSummarizer" extension tool in their browser and search for "coronavirus vaccine" on Google.

[0112] The extended tool uses the trained AIGC model to analyze search results in real time and extract key information, such as:

[0113] Keywords: new coronavirus, vaccine, effectiveness, safety, side effects, etc.

[0114] Entity information: mRNA vaccine, adenovirus vector vaccine, Pfizer, Moderna, etc.

[0115] Abstract: Briefly summarize the principles, effects and applicable populations of different types of vaccines.

[0116] Users can adjust analysis parameters according to personal preferences, such as:

[0117] Keyword weight: Setting “effectiveness” is more important than “side effects”.

[0118] Summary Length: Choose to generate a brief or detailed summary.

[0119] The extension tool dynamically adjusts the display of analysis results based on the user's historical behavior and preferences, for example:

[0120] Science enthusiasts: pay more attention to vaccine research progress and clinical trial data.

[0121] Ordinary users: pay more attention to the vaccination process and precautions.

[0122] Based on the same technical concept as the search method, the embodiment of the present invention also provides a text information search device, see Figure 3 , Figure 3 The structure diagram of the search device is as follows. The search device includes:

[0123] The acquisition module 301 is used to acquire the input text of the user performing the search task based on the Internet;

[0124] An acquisition module 302 is used to perform association expansion on the input text to obtain a text set of the input text, wherein the text set includes a plurality of keywords and / or sentences associated with the input text;

[0125] The training module 303 is used to train the preset AIGC model according to the text set and the user's demand search results until the AIGC model accurately outputs the demand search results corresponding to the input text;

[0126] The execution module 304 is used to execute the user's task to be searched according to the trained AIGC model to obtain the user's target search text for the task to be searched.

[0127] In an optional embodiment, the obtaining module includes:

[0128] The input acquisition submodule is used to input the input text into different search engines respectively to obtain the page search results of each search engine;

[0129] A sorting and obtaining submodule is used to prioritize all page search results to obtain at least one target search result;

[0130] A processing and obtaining submodule is used to perform word frequency statistics, part-of-speech tagging and vectorization processing on the target search results in sequence to obtain word vectors of keywords associated with the input text;

[0131] The first determination submodule is used to determine a set consisting of word vectors of all keywords as a text set.

[0132] In an optional embodiment, the ranking obtaining submodule includes:

[0133] An obtaining unit, used to obtain the priority of each page search result according to the number of outbound links and the number of inbound links represented by each page search result, wherein the number of outbound links is the number of other pages linked to the search result page, and the number of inbound links is the number of link jumps from the search result page to the seed page associated with the input text;

[0134] The determination unit is used to determine the page search results with a priority greater than the level threshold as target search results.

[0135] In an optional embodiment, the obtaining unit includes:

[0136] Get subunits for use according to the formula Get the priority P(U) of each page search result, where S(U) is the number of inbound links and L(U) is the number of outbound links.

[0137] In an optional embodiment, the training module includes:

[0138] A processing submodule, used for mapping the text collection and the demand search results to the same vector space, and calculating the similarity between the text collection and the demand search results;

[0139] A training adjustment submodule is used to train the AIGC model according to the similarity and adjust the model loss function based on the prediction results of the AIGC model;

[0140] The second determination submodule is used to determine that the AIGC model has completed training when the difference between the prediction result of the AIGC model and the demand search result is less than a preset condition.

[0141] In an optional embodiment, the searching device further includes:

[0142] The adjustment module is used to fine-tune the AIGC model according to the cross entropy loss function and the gradient descent algorithm to reduce the prediction error of the AIGC model.

[0143] In an optional embodiment, the execution module includes:

[0144] The compilation configuration submodule is used to compile the trained AIGC model into a WebAssembly module and configure it as a search plug-in for the user's browser;

[0145] The input and run submodule is used to input the search text of the task to be searched into the search plug-in and run the search plug-in;

[0146] The output acquisition submodule is used to obtain the user's target search text for the task to be searched based on the output result of the search plug-in.

[0147] Based on the same technical concept as the search method, an embodiment of the present invention also provides an electronic device, including a processor and a memory, wherein the memory is coupled to the processor and stores instructions, and when the instructions are executed by the processor, the electronic device executes the steps of any one of the search methods.

[0148] Based on the same technical concept as the search method, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of any one of the search methods are implemented.

[0149] The technical solution provided in the embodiments of the present invention has at least the following technical effects or advantages:

[0150] By obtaining the input text of the user performing a search task based on the Internet, the input text is associated and expanded to obtain a text set of the input text. Since the text set includes multiple keywords and / or sentences associated with the input text, the preset AIGC model can be trained through the text set and the user's demand search results. When the AIGC model accurately outputs the demand search results corresponding to the input text, it means that the AIGC model has been able to intelligently summarize and analyze the web page search results. The user's task to be searched is executed according to the trained AIGC model to obtain the user's target search text for the task to be searched. The technical solution of the present invention intelligently summarizes and analyzes web page search results according to the personalized needs of different types of users, can help different types of users quickly identify and extract key information in search results, and improves the efficiency and accuracy of text information search.

[0151] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0152] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (modules, systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0153] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0155] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0156] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A text information search method, characterized in that: The method comprises: Obtaining the input text of the user performing a search task based on the Internet; Performing association expansion on the input text to obtain a text set of the input text, wherein the text set includes a plurality of keywords and / or sentences associated with the input text; Training a preset AIGC model according to the text set and the user's demand search results until the AIGC model accurately outputs the demand search results corresponding to the input text; The user's task to be searched is executed according to the trained AIGC model to obtain the user's target search text for the task to be searched.

2. The text information search method according to claim 1, characterized in that: The associative expansion of the input text to obtain a text set of the input text includes: Inputting the input text into different search engines respectively to obtain page search results of each search engine; Prioritize all page search results to obtain at least one target search result; Performing word frequency statistics, part-of-speech tagging, and vectorization processing on the target search results in sequence to obtain word vectors of keywords associated with the input text; A set consisting of word vectors of all keywords is determined as the text set.

3. The text information search method according to claim 2, characterized in that: Prioritizing all page search results to obtain at least one target search result includes: Obtaining the priority of each page search result according to the number of outbound links and the number of inbound links represented by each page search result, wherein the number of outbound links is the number of other pages linked to the search result page, and the number of inbound links is the number of link jumps from the search result page to the seed page associated with the input text; The page search result with a priority greater than the level threshold is determined as the target search result.

4. The text information search method according to claim 3, characterized in that: The step of obtaining the priority of each page search result according to the number of outbound links and the number of inbound links represented by each page search result includes: According to the formula The priority P(U) of each page search result is obtained, where S(U) is the number of inbound links and L(U) is the number of outbound links.

5. The text information search method according to claim 1, characterized in that: The step of training a preset AIGC model according to the text set and the user's demand search results until the AIGC model accurately outputs the demand search results corresponding to the input text includes: Mapping the text set and the demand search results to the same vector space, and calculating the similarity between the text set and the demand search results; Training the AIGC model according to the similarity, and adjusting the model loss function based on the prediction result of the AIGC model; When the difference between the prediction result of the AIGC model and the demand search result is less than a preset condition, it is determined that the AIGC model has completed training.

6. The text information search method according to claim 5, characterized in that: Before determining that the AIGC model has completed training, the method further includes: The AIGC model is fine-tuned according to the cross entropy loss function and the gradient descent algorithm to reduce the prediction error of the AIGC model.

7. The text information search method according to claim 1, characterized in that: The executing the task to be searched by the user according to the trained AIGC model to obtain the target search text of the user for the task to be searched includes: Compile the trained AIGC model into a WebAssembly module and configure it as a search plug-in for the user's browser; Inputting the to-be-searched text of the to-be-searched task into the search plug-in, and running the search plug-in; According to the output result of the search plug-in, the target search text of the user for the task to be searched is obtained.

8. A text information search device, characterized in that: The device comprises: An acquisition module is used to acquire input text of a user performing a search task based on the Internet; An acquisition module, configured to perform association expansion on the input text to obtain a text set of the input text, wherein the text set includes a plurality of keywords and / or sentences associated with the input text; A training module, used to train a preset AIGC model according to the text set and the user's demand search results, until the AIGC model accurately outputs the demand search results corresponding to the input text; An execution module is used to execute the user's task to be searched according to the trained AIGC model to obtain the user's target search text for the task to be searched.

9. An electronic device, characterized in that: The electronic device comprises a processor and a memory, wherein the memory is coupled to the processor and stores instructions, and when the instructions are executed by the processor, the electronic device executes the steps of the method according to any one of claims 1 to 7.

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

Images