Searching method based on multi-source information fusion

Abstract

The invention discloses a search method based on multi-source information fusion, and relates to the technical field of multi-source information search, and the method comprises the steps: S1, obtaining a user operation instruction based on a redefined shortcut operation, so as to recognize a user demand; s2, acquiring search content of a user, and determining a search way for performing multi-source information search in combination with a natural language processing technology and a large language model; s3, searching based on the determined search way to obtain a corresponding search result; and S4, obtaining user preference information, integrating and analyzing the search result in combination with the user preference information to obtain an integration result, and returning the integration result to the user for viewing according to a set return rule. According to the method, the problem of low information search efficiency of the user in the mobile internet era is solved by providing a quick search way and automatically fusing and returning the multi-source data, meanwhile, a more accurate comprehensive data analysis result can be provided, and the search efficiency and the accuracy of data analysis are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-source information search, and in particular to a search method based on multi-source information fusion. Background Art

[0002] With the rapid growth of online information, the problems that come with it have gradually become prominent. In the PC era, the main way for users to search for Internet information was through search engines that were composed of web crawler technology that followed the Robots protocol. The information sources were of high depth and breadth, and users were required to confirm the search results one by one to get the desired results. After entering the mobile Internet era, various vertical industry application software APPs have flourished. Users have higher requirements for the diversity, immediacy and accuracy of search information, and the search method has expanded from traditional text search to multimodal search and analysis such as voice, photos, and articles. The entrance to information search has shifted from traditional search engines to the search entrances of multiple vertical APPs, and the results of information search have shifted from web pages to dynamics, products, posts, videos, etc., which has caused users' information search needs to become increasingly cumbersome and complex.

[0003] Currently, there are the following problems in information search on mobile terminals: (1) When searching for information, users need to switch back and forth between multiple apps, and a large amount of search result data needs to be manually integrated, screened and analyzed. As the number of apps increases, the time and energy cost of users' information search increases. (2) When using mobile terminal applications, users usually want to have a deeper understanding of the information they see. The traditional approach is to copy the information through the clipboard, exit the application and return to the desktop, find the browser, open the search website, paste the information, click search, obtain the search results, and manually organize the search result data. The information search process is very cumbersome, and if it is image information, the operation is even more complicated, resulting in a continuous decline in users' willingness to query and understand unknown information; (3) The search demand scenarios of users on mobile terminals are usually diverse. In order to ensure that the device can respond to the user's search needs more quickly in any scenario, traditional technical solutions provide a variety of methods, such as the more typical floating window button and voice wake-up. However, both methods have shortcomings. The floating window button will block the user's vision, affecting normal use and causing problems such as accidental touches. Voice wake-up will consume more power and requires the user to wake up by voice to respond, which cannot provide users with convenient and effective search services. (4) Although big model technology can now integrate public web page information on the Internet to provide comprehensive analysis results, big models cannot meet users' precise search needs well due to the real-time problems of pre-training data and the extensiveness of RAG (retrieval-augmented generation) data sources; (5) Traditional information search services are mainly provided by aggregating public information on the entire network, which is also the mainstream model of search engines and large language model knowledge questions and answers. Users obtain results by entering keywords or asking questions in natural language. Search engines and large language models use pre-collected and trained data to search, analyze, and integrate results, and finally generate search results. Although this method can help users acquire knowledge, it cannot fully combine the user's subjective intentions and behavioral habits as well as the user's private data for accurate retrieval. In order to obtain satisfactory search results, users often need to have an in-depth understanding and superb skills in the search keywords and prompt engineering technology, which is a great challenge for ordinary users. Summary of the invention

[0004] The present invention provides a search method based on multi-source information fusion to overcome the above technical problems existing in current mobile terminals when performing information search.

[0005] In order to achieve the above object, the technical solution of the present invention is: A search method based on multi-source information fusion, the specific steps include: S1: obtaining a user operation instruction based on the redefined quick operation to identify a user demand, wherein the user demand includes an information search demand; S2: obtaining a user search interface based on the user demand, obtaining user search content based on the user search interface, and combining natural language processing technology and a large language model to determine a search path for multi-source information search; The search approach includes a full-network information search and a robotic process automation search; The search scope of the whole network information search is several search platforms related to the user's search content; The search scope of the robotic process automation search is a number of application software related to the user's search content; S3: Search based on the determined search path to obtain corresponding search results; S4: Obtain user preference information, integrate and analyze the search results in combination with the user preference information to obtain an integrated result, and return the integrated result to the user for viewing according to a set return rule.

[0006] Furthermore, the specific steps of obtaining the user search content and combining the natural language processing technology and the large language model to determine the search path for multi-source information search include: 1) Based on natural language processing technology, determine whether the number of words in the user's search content exceeds the set threshold. If so, directly process the search content through the large language model and return the search results; otherwise, execute 2); 2) Use natural language processing technology to perform word segmentation and part-of-speech recognition on search content; 3) Automatically select the network-wide information search path or the robotic process automation search path based on word segmentation and part-of-speech recognition and the Bayesian algorithm. If the Bayesian algorithm cannot determine the search method, execute 4); The formula of the Bayesian algorithm is: P(A|B) = P(B|A)×P(A)P(B) in: P(A|B) represents the probability that the search content is the whole network search A when the user performs a certain search behavior B; P(B|A) represents the probability that the user actually performs a search behavior B if the user's search content is a full-network search A. P(A) represents the prior probability that the user's search content is a full-network search; P(B) represents the probability that a user actually performs a certain search behavior; 4) Use the large language model (LLM) to analyze the identified parts of speech, and select the full network information search path or the robotic process automation search path based on the analysis results, including: If the analysis result is a question-and-answer search, select the whole network information search method to obtain the search results; If the analysis result is a product search or a matching RPA script exists, select the Robotic Process Automation search path to obtain the search results.

[0007] Furthermore, the steps of implementing the robotic process automation search path include: Create several RPA collection scripts for launching application software; Match the corresponding RPA collection script based on the analysis results; The application software is started according to the matching RPA collection script, and a search is performed based on the started application software to obtain search results.

[0008] Furthermore, the user preference information includes the user's usage records, conversation records, user favorites, and personalized records set by the user in each application software.

[0009] Furthermore, the specific steps of obtaining user preference information, integrating and analyzing the search results in combination with the user preference information, and obtaining the integrated results include: Use the large language model to summarize and analyze the user's usage data according to the set period to obtain the user's preference information in each preference dimension; The large language model is set based on user preference information, including setting the context, rules and restrictions of the dialogue based on the system prompt, so as to integrate and analyze the search results to obtain integrated results.

[0010] Furthermore, the set return rules include: If the whole network information search method is adopted, the integrated results will be returned in the form of a floating window; If the Robotic Process Automation search approach is used, the integrated results are returned in the form of a dialogue window.

[0011] Furthermore, the redefined quick operation acquires the user operation instruction to identify the user demand, including: Redefine the functions of the existing volume up and down keys to use them as shortcut keys to start corresponding functions; Obtaining a function signal based on the redefined volume up and down key input, and performing a corresponding operation, including: if a single-click volume up key signal is obtained, performing a freeze screen operation, wherein the freeze screen operation includes covering the current application software interface with a floating window and displaying a clickable search function area of ​​the current application software interface; If a signal of clicking the volume down key is obtained, a quick search operation is performed, wherein the quick search operation includes starting a voice search, starting a photo search, or starting a text search; If a long press signal of the volume up or down button is obtained, the search interface is started and the voice recognition operation is performed synchronously. When it is recognized that the user releases the button, the voice recognition result is converted into text and the search is performed; If a double-click signal of the volume up or down key is obtained, the volume adjustment operation is performed.

[0012] Beneficial effects: The present invention obtains user operation instructions through redefined quick operations to identify user needs, obtain user search content, and combine natural language processing technology and large language models to determine the search path for multi-source information search; the search path includes full-network information search and robotic process automation search; the search scope of the full-network information search is several search platforms related to the user's search intention; the search scope of the robotic process automation search is several application software related to the user's search intention; obtain user preference information, integrate and analyze the search results in combination with the user preference information, obtain the integrated results, and return the integrated results to the user for viewing according to the set return rules. The present invention solves the problem of information search efficiency of users in the mobile Internet era by providing a quick search path and automatically integrating and returning multi-source data, effectively simplifies the cumbersome process of information search for users in the mobile Internet era, and takes into account the subjective intentions and behavioral habits of users, provides more accurate data sources through robotic process automation, and provides more accurate comprehensive data analysis results based on the user's search content and data collection results in combination with large language models, thereby greatly improving the search efficiency and the accuracy of data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0014] Figure 1 The present invention is a flowchart of a search method based on multi-source information fusion. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] This embodiment provides a search method based on multi-source information fusion, such as Figure 1 As shown, the specific steps include: S1: obtaining a user operation instruction based on the redefined quick operation to identify a user demand, wherein the user demand includes an information search demand; S2: obtaining a user search interface based on the user demand, obtaining user search content based on the user search interface, and combining natural language processing technology and a large language model to determine a search path for multi-source information search; The search approach includes a full-network information search and a robotic process automation search; The search scope of the whole network information search is several search platforms related to the user's search content; The search scope of the robotic process automation search is a number of application software related to the user's search content; S3: Search based on the determined search path to obtain corresponding search results; S4: Obtain user preference information, integrate and analyze the search results in combination with the user preference information to obtain an integrated result, and return the integrated result to the user for viewing according to a set return rule.

[0017] Specifically, this embodiment improves the traditional information search mode, greatly facilitating the user's search needs for any multimodal information, allowing the user to complete the search needs with one click, avoiding the user from performing tedious and repetitive operations, and can effectively integrate the search functions of various application software to achieve comprehensive aggregation and presentation of accurate information. It provides users with comprehensive data analysis results and decision-making recommendations through the large language model (LLM) technology, and can also assist users in conveniently recording information, and provide quick help functions, thereby comprehensively improving the user's search and information management efficiency.

[0018] In a specific embodiment, the redefined quick operation acquires the user operation instruction to identify the user demand, including: Redefine the functions of the existing volume up and down keys to use them as shortcut keys to start corresponding functions; Specifically, in this embodiment, preferably, the functions of the volume up and down keys of the mobile terminal using the Android system are redefined.

[0019] Get the function signal based on the redefined volume up and down key input and perform corresponding operations, including: If a signal of clicking the volume up button is obtained, a screen freeze operation is performed, wherein the screen freeze operation includes covering the current application software interface with a floating window and displaying a clickable search function area of ​​the current application software interface; If a signal of clicking the volume down key is obtained, a quick search operation is performed, wherein the quick search operation includes starting a voice search, starting a photo search, or starting a text search; If a long press signal of the volume up or down button is obtained, the search interface is started and the voice recognition operation is performed synchronously. When it is recognized that the user releases the button, the voice recognition result is converted into text and the search is performed; If a double-click signal of the volume up or down key is obtained, the volume adjustment operation is performed.

[0020] Specifically, to achieve the above functions, this embodiment uses the barrier-free service function of the mobile terminal to monitor the key events of the device, that is, by monitoring the on Key Event event of the barrier-free service; When the signal of the physical button is detected, it is determined whether the KEY_CODE of the on Key Event event is KEYCODE_VOLUME_UP or KEYCODE_VOLUME_DOWN. If so, the subsequent transmission process of the volume button to adjust the volume is interrupted, and the user's operation needs are distinguished and identified through the following formula: , in: Press_Type is the press type, including Long Press, Double Click, Short Press, and No Press; Indicates the timestamp of volume button pressing; Indicates the timestamp when the volume button is released; Indicates the time threshold for a short press; Indicates the time threshold for long press; Indicates the time interval threshold for double-clicking; The timestamp of the last key press, used for double-click detection; Indicates the timestamp of the last time a key was released, used for double-click detection.

[0021] Specifically, in view of the relatively low frequency of volume keys in daily use, and the fact that almost all mobile terminals currently only retain volume keys and power buttons, this embodiment rewrites the volume keys and freezes the screen interface operations so that users can go directly to search results with one click. At the same time, due to the restrictions on the underlying system permissions, the setting of shortcut keys is generally implemented through hardware and driver-level programs. In order to improve the operating efficiency on existing mobile devices, this embodiment chooses to use accessibility services to redefine the volume up and down keys, converting them into two physical shortcut keys with new functions, thereby providing users with a more convenient operating experience and keeping the original practicality of the volume key for adjusting the volume undisturbed. In practice, corresponding shortcut functions can also be designed according to the specific structure of the mobile terminal.

[0022] Specifically, by freezing the screen operation, the relevant information control node element clicked by the user is obtained, and the RPA (robotic process automation) + LLM comprehensive analysis results can be displayed in a floating window on the current page. In order to meet the more complex search requirements mentioned in the following steps, this embodiment can group the information selected by the user into data items of the same category as user preference data, so as to facilitate subsequent sorting and analysis using large language model technology.

[0023] In a specific embodiment, the specific steps of obtaining user search content and combining natural language processing technology and a large language model to determine a search path for multi-source information search include: 1) Based on natural language processing technology, determine whether the number of words in the user's search content exceeds the set threshold. If so, directly process the search content through the large language model and return the search results; otherwise, execute 2); Specifically, if the number of words in the search content exceeds the set threshold, it is judged as a long article. At this time, the search content is directly processed by the large language model and the search results are returned; 2) Use natural language processing technology to perform word segmentation and part-of-speech recognition on search content; 3) Automatically select the whole network information search path or the robot process automation search path based on word segmentation and part of speech recognition and using the Bayesian algorithm. If the search method cannot be determined by the Bayesian algorithm, execute 4); the formula of the Bayesian algorithm is: P(A|B) = P(B|A)×P(A)P(B) in: P(A|B) represents the probability that the search content is the whole network search A when the user performs a certain search behavior B; P(B|A) represents the probability that the user actually performs a search behavior B if the user's search content is a full-network search A. P(A) represents the prior probability that the user's search content is a full-network search; Specifically, P(A) needs to be estimated based on historical search data and user behavior habits; P(B) represents the probability that a user actually performs a certain search behavior; Specifically, P(B) needs to be estimated based on historical search data and user behavior habits.

[0024] Specifically, without considering other information, the probability P(A) that the user wants to use the whole network search method to search for content is estimated by the proportion of whole network searches in the user's historical search data. Based on the probability of a certain search behavior B, that is, regardless of the user's intention, P(B) is estimated by the total frequency of the search behavior in the historical search data; 4) Use the large language model (LLM) to analyze the identified parts of speech, and select the full network information search path or the robotic process automation search path based on the analysis results, including: If the analysis result is a question-and-answer search, select the whole network information search method to obtain the search results; If the analysis result is a product search or a matching RPA script exists, select the Robotic Process Automation search path to obtain the search results.

[0025] Specifically, this embodiment divides search paths into two categories according to user search content: one is the whole network information search path, which is used to obtain knowledge data and respond to search needs that do not require high real-time information; the other is the robotic process automation search path, which is used to provide RPA automatic collection of APP search data. The data collected through the two search paths must be subsequently comprehensively analyzed through language big model technology.

[0026] Specifically, when performing a search, the mobile terminal will prompt the user of the search path selected by the user. If the search path is not the search path required by the user, the user can manually interrupt and switch the search path.

[0027] In this embodiment, the user habits include the user's usage records, conversation records, collected and organized data, and the user personalization functions of each application connected to the RPA.

[0028] In a specific embodiment, the steps of implementing the robotic process automation search path include: Use the simulated click and text input functions of the accessibility service function to create several RPA collection scripts for launching application software; Match the corresponding RPA collection script based on the analysis results; Start the application software according to the matching RPA collection script, and search based on the started application software to obtain search results; Use a large language model to analyze and organize search results and return the search results.

[0029] Specifically, the RPA collection script can automatically open the search interface, automatically click the search box, enter the search content, click search, automatically collect data related to the search content, integrate the APP data of major vertical industries, and transmit it to the large language model.

[0030] Specifically, because the RPA collection script can submit the collected data to the large language model for unified analysis and organization, and give the final results, there is no need to perform structured preprocessing on the search results, which greatly reduces the difficulty of RPA script writing and subsequent data processing, improves the convenience of RPA script writing, and increases the stability of data collection. Through automated collection, the relevant APP data is collected and automatically integrated in sequence, effectively reducing the tedious complexity of users' vertical application searches, and combining the large language model to maximize the value of search result data. Finally, the collected keyword-related data is automatically aggregated into the data items of the current search task for subsequent data integration and analysis.

[0031] Specifically, when users search for specific products, they often need to frequently switch between multiple e-commerce apps in order to compare information such as product prices, descriptions, reviews, and delivery times from memory in a large amount of search result data. When users want to learn more about a restaurant, they usually need to search for a large amount of information such as related posts, reviews, locations, prices, and recommended dishes in multiple group-buying review-related apps, and decide whether to choose the restaurant based on their own judgment, which may not be comprehensive enough. This embodiment proposes a new search mode, multi-source information fusion search, which integrates search source information from different scenarios and user subjective intentions, behavioral habits, and private data, to further improve the accuracy and personalization of information search services, so that users can quickly obtain valuable information based on their subjective intentions in complex search results.

[0032] In a specific embodiment, the specific steps of obtaining user preference information, integrating and analyzing the search results in combination with the user preference information, and obtaining the integrated results include: Use the large language model to summarize and analyze the user's usage data according to the set period to obtain the user's preference information in each preference dimension; The large language model is set based on user preference information, including setting the context, rules and restrictions of the dialogue based on the system prompt, so as to integrate and analyze the search results to obtain integrated results.

[0033] Specifically, if the user is using the product for the first time, cold start recommendations can be performed based on the data processing capabilities of the large language model. Cold start recommendations are a method of making recommendations to users without a large amount of user data, such as using data such as user age, gender, and occupation to make recommendations.

[0034] In this embodiment, since the accessibility service function on the mobile terminal is unstable in processing page controls and needs access to common application data, the interface layout and data are converted into XML for processing by traversing the interface node elements in practice.

[0035] Specifically, in this embodiment, after the large language model uniformly analyzes and organizes all collected data and returns it, the user can continue to ask questions about the collected data, and the results returned by the large language model are the organized search results, and the user is allowed to click to view the original data.

[0036] In a specific embodiment, the robotic process automation search method further includes: The RPA recording function is built into the mobile terminal, allowing users to perform RPA recording based on the floating window freezing method. All applications installed on the mobile terminal are obtained through the get Installed Applications method of the Package Manager. When the user selects one of the applications, the mobile terminal system starts the application and builds a transparent floating window to cover the entire mobile terminal interface. The floating window TYPE is set to: TYPE_ACCESSIBILITY_OVERLAY, and the FLAG is set to: FLAG_LAYOUT_IN_SCREEN| FLAG_KEEP_SCREEN_ON | FLAG_NOT_TOUCHABLE | FLAG_NOT_TOUCH_MODAL | FLAG_NOT_FOCUSABLE | FLAG_FULLSCREEN, so as to process the user's click operation event and prevent the event from being transmitted to the target APP application.

[0037] Specifically, the floating window receives the user's click operation, obtains the corresponding control in the lower-level target APP application through the screen coordinates of the click, and obtains the unique identifier Id, Text, Bounds and Xpath of the control in the interface XML, as well as the operation type of the control, such as click, long press, slide, etc.; the click operation is recorded in the RPA script, and the floating window is hidden at the same time, the action is replayed, and the action is passed to the recorded APP, so that the user's operation is synchronized with the APP, and the user is provided with verification of the accuracy of the operation, and the above steps are repeated to record the user's operation path to gradually enter the target interface, and a movable menu is provided on the floating window to provide users with text input, data acquisition, homepage back, completion of recording, and cancellation of recording functions.

[0038] In this embodiment, in addition to allowing the user to click and identify the text content of the current page, the freeze screen function also allows the user to traverse the interface control elements to obtain scrollable controls (referring to data list controls, interface control nodes that need to be scrolled to load to obtain complete data), that is, the node where Accessibility NodeInfo.is Scrollable is equal to True, and the area of ​​the node is painted green to remind the user that the area can be scrolled to obtain complete data. The user slides on the area of ​​the frozen floating window, and the mobile terminal system identifies the sliding direction through the starting coordinates touched by the user, and calls AccessibilityNodeInfo.performAction in a loop to automatically slide the scroll control, and at the same time traverses all control nodes in the area whose attribute Text is not equal to empty to realize real-time collection of data in the area until the data is fully loaded or the user manually interrupts. The freeze screen scrolling function can realize the collection of data such as articles, web pages, transactions, comments, posts, etc. At the same time, it provides it to the large language model for processing, helping users to analyze and use mobile terminal content data better and faster.

[0039] Specifically, in terms of search entry, this embodiment introduces RPA technology and combines it with AcccessibilityService to implement fast and stable automated search tasks on mobile terminals, integrating search result data from various vertical industry apps.

[0040] Specifically, this embodiment provides great convenience and scalability for user-defined RPA to access private data sources by recording the user's operation path. Specifically, when the user starts the recording action, he first selects an APP application, and builds a transparent floating window while starting the APP. The floating window covers the entire mobile terminal interface and is used to record the user's clicks and input operations. When the user completes the recording of the task and saves the user's entire process, the user can access the entire RPA task search. Through the above operations, when executing the RPA search task, in addition to integrating the applications supported by the system, the user-defined RPA actions can also be replayed to meet the user's personalized needs.

[0041] Specifically, this embodiment allows users to build prompts according to customized requirements to achieve more comprehensive data mining. At the same time, it makes full use of the existing mobile terminal hardware conditions, uses Android auxiliary functions to rewrite volume button events and combines them with freeze screen operations to achieve one-click intelligent search in multiple scenarios, ensuring that users can quickly reach search results in any situation.

[0042] In this embodiment, for APPs that do not have RPA scripts set, RPA scripts are generated by recording user operations, simulating user operations, and gradually entering the data interface to obtain all data in the interface. Finally, access to the APP data is achieved. Through automated collection, relevant APP data is collected and automatically integrated in turn, which effectively reduces the cumbersome complexity of users' vertical application searches, and combines the large language model to maximize the value of search result data. Finally, the collected keyword-related data is automatically aggregated into the data items of the current search task for subsequent data integration and analysis.

[0043] In a specific embodiment, the set return rule includes: If the whole network information search method is adopted, the integrated results will be returned in the form of a floating window; Specifically, the TYPE_ACCESSIBILITY_OVERLAY permission is used to create an accessible floating window, so that search results can be displayed on any interface, and users no longer need to apply for other floating permissions; Specifically, in the traditional method, when users search for interface content, they often return to the search start interface or repeatedly compare with the results. In order to avoid users from repeatedly switching between the search results page and the start page, this embodiment displays the search results page in the form of a floating window, so that the search results are displayed where the user initiates the search, effectively improving the user's willingness to search for information.

[0044] For example: when a user needs to search for an unfamiliar word on the interface, he only needs to click the volume up button and click the word to get detailed information about the word.

[0045] If the Robotic Process Automation search approach is used, the integrated results are returned in the form of a dialogue window.

[0046] Specifically, in the robotic process automation search approach, after collecting user search-related data through robotic process automation and large language models, it is analyzed based on the user's search intent, and combined with the user's historical data to mine and analyze preferences, habits, behaviors, etc., to provide users with a series of knowledge questions and answers, decision-making suggestions, creation generation, text summarization, language translation, report generation, etc., and provide users with question guidance to bring more practical application scenarios to the search results.

[0047] For example, when a user searches for hotel information in a certain place, the system automatically integrates all relevant hotel information in that place and provides users with suitable hotel recommendations such as price, evaluation, location, etc. Or when a user needs to learn more about a piece of news on the interface, with just one click, the system can automatically search for all relevant information about the news and provide a text summary.

[0048] In a specific embodiment, the data collection of the whole network information search is deeply accessed and uses crawler technology to efficiently integrate the massive data resources disclosed by major network platforms. At the same time, the advanced technology and capabilities of the large language model are integrated to analyze and organize the keywords of the user's search needs, and accurately capture and filter the information highly related to these keywords in the network data. The whole network information search not only relies on the wide coverage and rapid indexing of data, but also makes full use of the advantages of the large language model in understanding natural language, identifying semantic relationships, and generating accurate replies, thereby greatly improving the accuracy and relevance of search results, and providing users with a richer and more comprehensive information retrieval method.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A search method based on multi-source information fusion, characterized in that: The specific steps include: S1: obtaining a user operation instruction based on the redefined quick operation to identify a user demand, wherein the user demand includes an information search demand; S2: obtaining a user search interface based on the user demand, obtaining user search content based on the user search interface, and combining natural language processing technology and a large language model to determine a search path for multi-source information search, the specific steps include: 1) Based on natural language processing technology, determine whether the number of words in the user's search content exceeds the set threshold. If so, directly process the search content through the large language model and return the search results; otherwise, execute 2); 2) Use natural language processing technology to perform word segmentation and part-of-speech recognition on search content; 3) Automatically select the network-wide information search path or the robotic process automation search path based on word segmentation and part-of-speech recognition and the Bayesian algorithm. If the Bayesian algorithm cannot determine the search method, execute 4); 4) Use a large language model to analyze the identified parts of speech, and select a network-wide information search path or a robotic process automation search path based on the analysis results; The search scope of the whole network information search is several search platforms related to the user's search content; The search scope of the robotic process automation search is a number of application software related to the user's search content; S3: Search based on the determined search path to obtain corresponding search results; S4: Obtain user preference information, integrate and analyze the search results in combination with the user preference information to obtain an integrated result, and return the integrated result to the user for viewing according to a set return rule.

2. The search method based on multi-source information fusion according to claim 1, characterized in that: The formula of the Bayesian algorithm is: P(A|B) = P(B|A)×P(A)P(B) in: P(A|B) represents the probability that the search content is the whole network search A when the user performs a certain search behavior B; P(B|A) represents the probability that the user actually performs a search behavior B if the user's search content is a full-network search A. P(A) represents the prior probability that the user's search content is a full-network search; P(B) represents the probability that a user actually performs a certain search behavior.

3. The search method based on multi-source information fusion according to claim 2 is characterized in that: Use a large language model to analyze the identified parts of speech, and select a full-network information search path or a robotic process automation search path based on the analysis results, including: If the analysis result is a question-and-answer search, select the whole network information search method to obtain the search results; If the analysis result is a product search or a matching RPA script exists, select the Robotic Process Automation search path to obtain the search results.

4. The search method based on multi-source information fusion according to claim 3 is characterized in that: The steps of implementing the robotic process automation search approach include: Create several RPA collection scripts for launching application software; Match the corresponding RPA collection script based on the analysis results; The application software is started according to the matching RPA collection script, and a search is performed based on the started application software to obtain search results.

5. The search method based on multi-source information fusion according to claim 4 is characterized in that: The user preference information includes the user's usage records, conversation records, user favorites, and personalized records set by the user in each application software.

6. The search method based on multi-source information fusion according to claim 5, characterized in that: The specific steps of obtaining user preference information, integrating and analyzing the search results in combination with the user preference information, and obtaining the integrated results include: Use the large language model to summarize and analyze the user's usage data according to the set period to obtain the user's preference information in each preference dimension; The large language model is set based on user preference information, including setting the context, rules and restrictions of the dialogue based on the system prompt, so as to integrate and analyze the search results to obtain integrated results.

7. The search method based on multi-source information fusion according to claim 6, characterized in that: The set return rules include: If the whole network information search method is adopted, the integrated results will be returned in the form of a floating window; If the Robotic Process Automation search approach is used, the integrated results are returned in the form of a dialogue window.

8. The search method based on multi-source information fusion according to claim 7, characterized in that: The redefined quick operation acquires the user operation instruction to identify the user demand, including: Redefine the functions of the existing volume up and down keys to use them as shortcut keys to start corresponding functions; Obtaining a function signal based on the redefined volume up and down key input, and performing a corresponding operation, including: if a single-click volume up key signal is obtained, performing a freeze screen operation, wherein the freeze screen operation includes covering the current application software interface with a floating window and displaying a clickable search function area of ​​the current application software interface; If a signal of clicking the volume down key is obtained, a quick search operation is performed, wherein the quick search operation includes starting a voice search, starting a photo search, or starting a text search; If a signal of long pressing the volume up or down button is obtained, the search interface is started and the voice recognition operation is performed synchronously. When it is recognized that the user releases the button, the voice recognition result is converted into text and the search is performed; If a double-click signal of the volume up or down key is obtained, the volume adjustment operation is performed.

Images