Information collection device, information collection method, and program

Abstract

To provide an information collection device, an information collection method, and a program capable of collecting accurate information without performing detailed adjustment.SOLUTION: The information collection device performs learning of a classifier model on the basis of labeling data indicating whether any web page is related to an investigation purpose by a logical value, determines whether information related to the investigation purpose is described in an input web page on the basis of a learning result of the classifier model, classifies the input web page, instructs a generative AI to extract information related to the investigation purpose from the web page classified as describing information related to the investigation purpose, and stores the information related to the investigation purpose extracted by the generative AI.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to an information collection device, an information collection method, and a program. [Background technology]

[0002] In recent years, it has become possible to collect vast amounts of information through various means, including websites, and the information industry, which handles information, is making a significant contribution to the development of the economy and companies. As a result, there is a great demand for information collection and processing technology, and there is a need for technology that can accurately collect vast amounts of information from a variety of sources and constantly update it.

[0003] For example, Patent Document 1 proposes a technology for a chatbot management support system and method that compares possible answers to a question, identifies an answer with a certainty level above a reference value, and responds. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2021-21990 Summary of the Invention [Problem to be solved by the invention]

[0005] Meanwhile, in the technology for collecting and outputting information from web pages, there has been a demand for technology that can collect accurate information without detailed adjustments, even for targets where the structure of the web page that provides the information is indefinite and the information is frequently updated.

[0006] An object of the present disclosure is to provide an information collection device, an information collection method, and a program that solve the above-mentioned problems. [Means for solving the problem]

[0007] An information collection device according to one embodiment of the present disclosure includes a learning unit that trains a classifier model to determine whether an input web page contains information related to the research purpose based on labeling data that indicates, using a logical value, whether a given web page is related to the research purpose; a web page classification unit that determines, based on the learning results of the classifier model, whether the input web page contains information related to the research purpose and classifies the input web page based on the determination results; a research execution unit that instructs a generation AI to extract information related to the research purpose from web pages that are classified by the web page classification unit as containing information related to the research purpose among the input web pages; and a database that stores information related to the research purpose extracted by the generation AI.

[0008] An information collection method according to one embodiment of the present disclosure includes the steps of: training a classifier model to determine whether an input web page contains information related to a research purpose based on labeling data indicating, as a logical value, whether a given web page is related to a research purpose; classifying the input web page based on the training results of the classifier model based on whether the input web page contains information related to the research purpose; instructing a generation AI to extract information related to the research purpose from web pages classified as containing information related to the research purpose among the input web pages; and storing the information related to the research purpose extracted by the generation AI in a database.

[0009] A program according to one embodiment of the present disclosure causes a computer of an information collection device to function as a learning function that trains a classifier model to determine whether an input web page contains information related to the research purpose based on labeling data that indicates, as a logical value, whether a given web page is related to the research purpose; a web page classification function that classifies the input web page based on the learning results of the classifier model and whether the input web page contains information related to the research purpose; a research execution function that instructs a generation AI to extract information related to the research purpose from web pages classified as containing information related to the research purpose among the input web pages; and a storage function that stores the information related to the research purpose extracted by the generation AI in a database. [Effects of the Invention]

[0010] According to the above aspect, accurate information can be collected without detailed adjustments even for a target whose information source has an indefinite structure and whose information is frequently updated. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a block diagram illustrating a configuration of an information collection device according to the present disclosure. [Figure 2] 10 is a flowchart for explaining the operation of a survey execution unit of the information processing device according to the present disclosure. [Figure 3] 10 is a flowchart illustrating the operation of a learning unit in the information collection device according to the present disclosure. [Figure 4] FIG. 10 is a conceptual diagram illustrating an example of the configuration of labeling data by the information collection device according to the present disclosure. [Figure 5] 10 is a flowchart for explaining the operation of a survey execution unit of the information processing device according to the present disclosure. [Figure 6] FIG. 10 is a conceptual diagram illustrating an example of the configuration of homepage data stored in the information collection device according to the present disclosure. [Figure 7]10 is a flowchart illustrating an operation of a web page classification unit in the information processing device according to the present disclosure. [Figure 8] FIG. 10 is a conceptual diagram illustrating an example of the configuration of data extracted by the information collection device according to the present disclosure. [Figure 9] 10 is a conceptual diagram illustrating an example of the configuration of survey data obtained by the information collection device according to the present disclosure. [Figure 10] FIG. 10 is a conceptual diagram showing an example of survey result data by related generation AI (Artificial Intelligence) for comparison with survey result data by the information collection device according to the present disclosure. [Figure 11] FIG. 10 is a conceptual diagram illustrating an example of survey result data obtained by the information processing device according to the present disclosure. [Figure 12] FIG. 10 is a block diagram illustrating another example configuration of the information collection device according to the present disclosure. [Figure 13] 10 is a flowchart for explaining the operation of a survey execution unit of the information processing device according to the present disclosure. [Figure 14] FIG. 10 is a block diagram illustrating another example configuration of the information collection device according to the present disclosure. [Figure 15] FIG. 1 is a hardware configuration diagram of an information processing device according to the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, each embodiment will be described with reference to the drawings. In all drawings, the same or corresponding components are denoted by the same reference numerals, and common descriptions will be omitted.

[0013] First Embodiment Hereinafter, an embodiment according to the present disclosure will be described with reference to the drawings. Fig. 1 is a block diagram showing the configuration of an information collection device according to the present disclosure. In Fig. 1, the information collection device 1 comprises a server / container 10 and a database / model 20. The server / container 10 comprises a scheduler 11, a process 12, and a module 13. The process 12 comprises a survey execution unit 121, and the module 13 comprises a learning unit 131 and a web page classification unit 132. The database / model 20 comprises a database 21 and a classifier model 22. The database 21 includes labeling data 211, homepage data 212, extracted data 213, and survey result data 214.

[0014] The scheduler 11 periodically executes the process 12 at an arbitrary frequency. In response to an execution request from the scheduler 11, the investigation execution unit 121 calls the module 13 and receives the processing result from the module 13. More specifically, the investigation execution unit 121 refers to the labeling data 211 to check whether there is unlearned data that has not been used to train the classifier model 22, and if unlearned data is included, calls the learning unit 131 to train the classifier model 22. The investigation execution unit 121 also refers to the homepage data 212 to search for hyperlinks from the homepage to be investigated, and executes the following processing while recursively accessing descendant pages (branches linked to the homepage).

[0015] The survey execution unit 121 inputs each web page to the web page classification unit 132 and receives the classification results from the web page classification unit 132. Based on the classification results, the survey execution unit 121 determines whether the input web page has been classified as relevant to the survey purpose. If the web page has been classified as relevant to the survey purpose, the survey execution unit 121 calls a generation AI (Artificial Intelligence) 30 to extract information relevant to the survey purpose from the web page and writes the response result of the generation AI 30 to the extracted data 213. After completing the above-mentioned processing for all websites, the survey execution unit 121 finally evaluates the data by referring to the extracted data 213 and writes data whose evaluation value meets the reference value to the survey result data 214.

[0016] The learning unit 131 and web page classification unit 132 in module 13 process information received from the research execution unit 121 and return a response to the research execution unit 121. More specifically, in response to a call from the research execution unit 121, the learning unit 131 refers to the labeling data 211, extracts text from each web page, converts it into a vector, and uses the vector and the label associated with the web page to train the classifier model 22. The web page classification unit 132 extracts text from the input web page, converts it into a vector, classifies the converted vector using the classifier model 22 based on whether it is relevant to the research purpose, and returns the classification result to the research execution unit 121.

[0017] In the database 21, the labeling data 211 is data created in advance by the user and indicates, using a logical value (label; True or False) whether a given web page (homepage) is relevant to the research objective. The homepage data 212 is data consisting of a list of uniform resource locators (URLs) of homepages of websites referenced as research targets. The information collection device 1 requires a certain number of labeling data 211 and homepage data 212 to operate. The extracted data 213 is a response result including information on the research objective extracted by the generation AI 30. The research result data 214 is final research result data for which the research execution unit 121 determines that the evaluation value satisfies the reference value. The classifier model 22 is, for example, a learning result obtained by a machine learning algorithm to classify vectors converted from text extracted from web pages by the web page classification unit 132 based on whether the vectors are relevant to the research objective. The classifier model 22 may use a technology called support vector machine (SVM), which handles classification and regression using supervised learning.

[0018] The generation AI 30 may be any external generation AI. When the generation AI 30 receives an instruction from the survey execution unit 121 to extract information related to the survey purpose from a web page, the generation AI 30 extracts the information related to the survey purpose based on the link of the given web page and returns the information to the survey execution unit 121.

[0019] 2 is a flowchart for explaining the operation of the investigation execution unit of the information processing device according to the present disclosure. When the investigation execution unit 121 receives a periodic execution request from the scheduler 11 or an execution instruction from a user, it first refers to the labeling data 211 and determines whether there is unlearned data that has not been used to train the classifier model 22 (step S10). If there is no unlearned data (NO in step S10), the process ends. On the other hand, if there is unlearned data (YES in step S10), the learning unit 131 is called (step S12).

[0020] 3 is a flowchart illustrating the operation of the learning unit in the information collection device according to the present disclosure. When called by the investigation execution unit 121, the learning unit 131 extracts text from each web page (step S20) and converts each text into a vector (step S22). For example, the learning unit 131 may use a technique called Term Frequency·Inverse Document Frequency (TF·IDF) to convert the text into a vector. Next, the learning unit 131 trains the classifier model 22 using the vectors and labels associated with the web pages (step S24).

[0021] FIG. 4 is a conceptual diagram illustrating an example of the configuration of labeling data generated by an information collection device according to the present disclosure. The labeling data 211 is data input in advance by a user. As shown in FIG. 3, the labeling data 211 is composed of the URL of a web page to be collected and a logical value (label: True, False) indicating whether the web page contains information to be collected as a research target. The labeling data 211 shown in FIG. 4 is an example of checking the support expiration date listed in the release notes (information regarding the support expiration date) of OSS (Open Source Software). The labeling data 211 includes the web page (URL) of the target site, and if the web page is a research target, it is set to "True" and if not, it is set to "False." In the illustrated example, "http: / / xaaaa.org / abc.htlm" is set to "True," "http: / / xaaaa.org / def.htlm" is set to "False," "http: / / www.xbbbb.com / support / jkl.html" is set to "True," and "http: / / www.xbbbb.com / def / " is set to "False."

[0022] 5 is a flowchart for explaining the operation of the investigation execution unit of the information processing device according to the present disclosure. When an execution request is received from scheduler 11 or an execution instruction is received from a user, investigation execution unit 121 refers to homepage data 212 to search for hyperlinks from the homepage to be investigated, and executes the following process while recursively accessing descendant pages (branches linked to the homepage) (step S30; loop 1, S32; loop 2). First, investigation execution unit 121 inputs each web page in homepage data 212 to web page classification unit 132 (step S34).

[0023] 6 is a conceptual diagram showing an example of the configuration of homepage data collected by the information collection device according to the present disclosure. The homepage data 212 is data indicating the homepage URLs of the official websites of each OSS being investigated. In the illustrated example, from top to bottom, they are "http: / / xaaaa.org / ," "http: / / www.xbbbb.com / ," "http: / / xcccc.org / ," and "http: / / www.xdddd.com / ."

[0024] 7 is a flowchart illustrating the operation of the web page classification unit of the information processing device according to the present disclosure. When each web page is input from the research execution unit 121, the web page classification unit 132 extracts text from the input web page (step S50). Next, the web page classification unit 132 converts the text into vectors (step S52) and classifies the converted vectors using the classifier model 22 based on whether they are related to the research purpose (step S54). Thereafter, the web page classification unit 132 returns the classification results to the research execution unit 121 (step S56) and ends the process.

[0025] Returning to FIG. 5, the investigation execution unit 121 determines whether the input web page is classified as a web page related to the investigation purpose based on the classification result from the web page classification unit 132 (step S36). If the input web page is classified as a web page related to the investigation purpose (YES in step S36), the investigation execution unit 121 instructs the generation AI 30 to extract information related to the investigation purpose from the web page (step S38). The investigation execution unit 121 provides the generation AI 30 with a link to the web page and instructs it to extract information related to the investigation purpose (for example, if the information is the support expiration date of an OSS, extract the URL information of the OSS from this web page for each version). Since the investigation execution unit 121 instructs the operation itself rather than specific instructions (such as a time), it is easier to obtain relatively accurate information. Furthermore, since the labeling data 211 is divided into True / False, it is easier to obtain accurate information. The investigation execution unit 121 then writes the response result of the generation AI 30 to the extraction data 213 (step S40).

[0026] Next, the investigation execution unit 121 determines whether the number of detections is equal to or greater than a predetermined value (step S42). If the number is not equal to or greater than the predetermined value (NO in step S42), the process returns to step S38 and repeats the above-described process. That is, extraction by the generation AI 30 is repeated any number of times for one web page. This is to later evaluate whether the generation AI 30 is consistently returning similar answers.

[0027] FIG. 8 is a conceptual diagram showing an example of the configuration of data extracted by the information collection device according to the present disclosure. The example shown in FIG. 8 shows extracted data 213 when checking the support expiration date of an open source. In this case, the extracted data 213 consists of the URL of the source web page, the last update date, the number of attempts, the OSS, and the EoL (End of Life). In the example shown, for the source "http: / / xaaaa.org / xaa-α.html", the last update date "2017 / 6 / 30" obtained as the support expiration date, the number of attempts "1" indicating the number of times the same answer (True or False) was obtained from the generation AI 30, the OSS "OSSα" indicating the release name, and the EoL "True" indicating whether the support expiration date has arrived. The number of lines is "2", the OSS is "OSSα", and the EoL is "True". For the source "http: / / xaaaa.org / xaa-β.html", the last update date is "2017 / 12 / 13", the number of attempts is "1", the OSS is "OSSβ", and the EoL is "False". For the source "http: / / xaaaa.org / xaa-β.html", the last update date is "2017 / 12 / 13", the number of attempts is "2", the OSS is "OSSβ", and the EoL indicating whether or not the support period has expired is "False".

[0028] On the other hand, if the input web page is not classified as a web page related to the research purpose (NO in step S36), the research execution unit 121 skips the above-mentioned processes (steps S38 to S42), returns to step S32 (loop 2), recursively accesses descendant pages (branches linked to homepages), and repeats the above-mentioned processes. Also, if the number of detections exceeds a predetermined value (YES in step S42), the research execution unit 121 returns to step S32 (loop 2), recursively accesses descendant pages (branches linked to homepages), and repeats the above-mentioned processes. Then, when loop 2 ends, the research execution unit 121 returns to step S30 (loop 1), repeats the same processes, and performs the above-mentioned processes for all homepages under research.

[0029] After that, when the above-mentioned processing is completed for all websites (loops 1 and 2 are completed), the survey execution unit 121 finally evaluates the data by referring to the extracted data 213, selects data whose evaluation value meets the standard value (highly credible data), and writes it to the survey result data 214 (step S44). The evaluation value is determined mainly based on the number of similar data, the number of contradictory data, and the last update date of the reference source, but can be set appropriately depending on the information to be surveyed, and weighting may also be performed. Then, the processing ends.

[0030] 9 is a conceptual diagram showing an example of the configuration of investigation result data by the information collection device according to the present disclosure. The investigation result data 214 finally obtained consists of data whose evaluation value meets the reference value (highly reliable data). In the example shown in FIG. 9, it consists of OSS and its EoL (True or False), with EoL "True" for OSS "Tomcat 8.0", EoL "True" for OSS "Tomcat 8.5", EoL "False" for OSS "Tomcat 9.0", and EoL "False" for OSS "Tomcat 10.0".

[0031] If the generation AI is simply instructed by providing web page information, there is a high possibility that it will use unreliable websites / information from unknown sources, or generate information that is not based on facts / information that is insufficient because the amount of information referenced is too large. In contrast, the information collection device 1 disclosed herein sets labeling data 211 in advance, which labels each web page as to whether or not it contains information related to a specific research purpose, classifies web pages from specified websites that contain information related to the research purpose, and instructs the generation AI 30 to extract the necessary information from the classified web pages. This reduces the possibility of hallucination and enables the collection of more accurate information related to the research purpose.

[0032] FIG. 10 is a conceptual diagram showing an example of survey result data generated by a related generation AI for comparison with survey result data generated by the information collection device according to the present disclosure. FIG. 11 is a conceptual diagram showing an example of survey result data generated by the information collection device 1 according to the present disclosure. Consider an example in which information is obtained for each version of open source software (OSS) being surveyed regarding whether sales and support for the OSS have ended (EOL). Because each OSS is developed by a different developer, the structure of the official website also varies significantly depending on the OSS, making it difficult to collect information through web scraping. Furthermore, because OSS versions are frequently updated, information on websites other than the official website is likely to be inaccurate.

[0033] The example shown in Figure 10 shows the result of instructing a generation AI chatbot "A" to list the names of OSS supported by Company B's OSS support service C. The instruction is "Please list the names of OSS supported by Company B's OSS support service C." In the response given by generation AI chatbot "A" to this instruction, black text without hatching (diagonal lines) indicates OSS that is actually supported, and blue text with hatching (diagonal lines) indicates OSS that is not actually supported. In this way, generation AI 30 may generate information that is contrary to the facts.

[0034] On the other hand, in the information collection device 1 according to the present disclosure, as shown in Fig. 11, the result of adding an instruction to refer only to the information on the specified web page to the question entered in Fig. 10 is shown. All of the OSS names provided in the response are actually supported OSS, and it can be seen that hallucination can be suppressed by limiting the information to be referred to.

[0035] In the information collection device 1 of this embodiment, labeling data 211 is set in advance to label each web page as to whether or not it contains information related to a specific research purpose, and the generation AI 30 is instructed to classify web pages from a specified website that contain information related to the research purpose and extract the necessary information from the classified web pages, thereby reducing the possibility of hallucination occurring and collecting more accurate information related to the research purpose.

[0036] In addition, the extracted information is evaluated according to predetermined conditions to select highly credible information, further reducing the possibility of outputting information that is not based on facts. Furthermore, even for information whose source web page structure is indefinite and frequently updated (e.g., end-of-life information for OSS), automatic information collection can be achieved without detailed adjustments. Furthermore, because the classifier model 22 is configured to exclude clearly irrelevant web pages, the number of processing times of the generation AI, which would otherwise take a huge amount of time, can be significantly reduced compared to processing all web pages without using the classifier model 22.

[0037] Second Embodiment Hereinafter, an embodiment according to the present disclosure will be described with reference to the drawings. FIG. 12 is a block diagram showing another example configuration of an information collection device according to the present disclosure. In FIG. 12, database 21 of database / model 20 includes a data group called crawling data 215 in addition to labeling data 211, homepage data 212, extracted data 213, and survey result data 214. Crawling data 215 is data collected from websites by a "crawler" and is used to determine whether or not there are differences (changes) in a web page since the previous survey, and if there are no differences, to exclude the web page from the target of information collection. Furthermore, ChatGPT (registered trademark), an artificial intelligence chatbot, may be used as generation AI 30, for example.

[0038] 13 is a flowchart for explaining the operation of the investigation execution unit of the information processing device according to the present disclosure. Note that the learning process of the classifier model 22 by the learning unit 131 is the same as the flowcharts shown in FIGS. 2 and 3 of the first embodiment described above, and therefore a description thereof will be omitted.

[0039] When the survey execution unit 121 receives a periodic execution request from the scheduler 11 or an execution instruction from a user, it first refers to the homepage data 212 to search for hyperlinks from the homepage to be surveyed, and then performs the following processing while recursively accessing descendant pages (branches linked to the homepage) (step S60; loop 1, S62; loop 2). The survey execution unit 121 first refers to the crawling data 215 and checks whether there are any differences between each web page and the previous time the survey was performed (step S64). If there are any differences (YES in step S64), the survey execution unit 121 replicates (crawls) the web page (step S66). The crawled data is written to the crawling data 215. On the other hand, if there are no differences (NO in step S64), the survey execution unit 121 skips all subsequent processing and repeats loop 2.

[0040] Next, the research execution unit 121 inputs each web page of the homepage data 212 to the web page classification unit 132 (step S68). When each web page is input from the research execution unit 121, the web page classification unit 132 executes the flowchart shown in FIG. 7 of the first embodiment described above. Specifically, the web page classification unit 132 extracts text from the input web page and converts it into a vector, classifies the converted vector using the classifier model 22 based on whether it is relevant to the research purpose (for example, whether it is an OSS release note), and returns the classification result to the research execution unit 121.

[0041] Next, the investigation execution unit 121 determines whether the input web page is classified as a web page related to the investigation purpose based on the classification result from the web page classification unit 132 (step S70). If the input web page is classified as a web page related to the investigation purpose (YES in step S70), the investigation execution unit 121 instructs the generation AI 30 to extract information related to the investigation purpose from the web page (step S72). Specifically, the investigation execution unit 121 provides the generation AI 30 with a link to the web page and instructs it to extract the information related to the investigation purpose (for example, if the information is the support expiration date of an OSS, extract the URL information of the OSS from this web page for each version). Since the investigation execution unit 121 instructs the operation itself rather than specific instructions (such as the time), it is easier to obtain relatively accurate information. Furthermore, since the labeling data 211 is divided into True / False, it is easier to obtain accurate information. Then, the investigation execution unit 121 writes the response result of the generation AI 30 to the extraction data 213 (step S74).

[0042] Next, the investigation execution unit 121 determines whether the number of detections is equal to or greater than a predetermined value (step S76). If the number is not equal to or greater than the predetermined value (NO in step S76), the process returns to step S72 and repeats the above-described process. That is, extraction by the generation AI 30 is repeated any number of times for one web page. This is to later evaluate whether the generation AI 30 is consistently returning similar answers.

[0043] On the other hand, if the input web page is not classified as a web page related to the research purpose (NO in step S70), the research execution unit 121 skips the above-mentioned processes (steps S72 to S76), returns to step S62 (loop 2), recursively accesses descendant pages (branches linked to homepages), and repeats the above-mentioned processes. Also, if the number of detections exceeds a predetermined value (YES in step S76), the research execution unit 121 returns to step S62 (loop 2), recursively accesses descendant pages (branches linked to homepages), and repeats the above-mentioned processes. Then, when loop 2 ends, the research execution unit 121 returns to step S60 (loop 1), repeats the same processes, and performs the above-mentioned processes for all homepages under research.

[0044] After that, when the above-mentioned process is completed for all websites (loops 1 and 2 are completed), the survey execution unit 121 finally evaluates the data by referring to the extracted data 213, selects data whose evaluation value meets the standard value (highly credible data), and writes the selected data to the survey result data 214 (step S78). Then, the process ends.

[0045] In the information collection device 1 of this embodiment, the crawling data 215 is referenced to check whether there is a difference between the last time each web page was investigated, and if there is a difference, the crawling data 215 is updated, and if there is no difference, all information extraction from the web page is skipped. Therefore, in addition to the effects of the first embodiment described above, the number of processes of the generation AI, which requires a huge processing time, can be significantly reduced, and the overall processing time can be shortened.

[0046] Third Embodiment Hereinafter, an embodiment according to the present disclosure will be described with reference to the drawings. 14 is a block diagram showing another example configuration of an information collection device according to the present disclosure. The information collection device 100 includes: a learning unit 103 that trains a classifier model 102 to determine whether an input web page describes information related to the research purpose based on labeling data 101 that indicates, using a logical value, whether a given web page is related to the research purpose; a web page classification unit 104 that determines, based on the learning result of the classifier model 102, whether the input web page describes information related to the research purpose and classifies the input web page based on the determination result; a search execution unit 106 that instructs a generation AI 105 to extract information related to the research purpose from web pages that are classified by the web page classification unit 104 as including information related to the research purpose among the input web pages; and a database 107 that stores the information related to the research purpose extracted by the generation AI 105.

[0047] The information collection device 100 of this embodiment sets labeling data 101 in advance, which labels each web page to indicate whether it contains information related to a specific research purpose.The web page classification unit 104 classifies web pages from a specified website that contain information related to the research purpose, and the research execution unit 106 instructs the generation AI 30 to extract the necessary information from the classified web pages.This reduces the possibility of hallucination occurring and makes it possible to collect more accurate information related to the research purpose.

[0048] In the above-described embodiment, it is conceivable that an arbitrary number of multiple web pages may be input together into the generation AI 30 to shorten the processing time. Furthermore, in the above-described embodiment, if the classifier model 22 makes a classification error and the generation AI 30 determines that the web page does not contain information relevant to the research purpose, it is possible to improve the accuracy of the classifier model 22 by writing data that the web page is not relevant to the research purpose into the labeling data 211 and allowing it to perform self-learning.

[0049] FIG. 15 is a hardware configuration diagram of an information collection device according to the present disclosure. As shown in FIG. 15, the information collection device 1 is a computer including various hardware components such as a CPU (Central Processing Unit) 201, a ROM (Read Only Memory) 202, a RAM (Random Access Memory) 203, a database 204, a communication module 205, and the like.

[0050] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate.

[0051] Some or all of the above-described embodiments can be described as, but are not limited to, the following supplementary notes.

[0052] (Appendix 1) a learning unit that learns a classifier model to determine whether an input web page contains information related to the research purpose, based on labeling data that indicates, using a logical value, whether a given web page is related to the research purpose; and a web page classification unit that determines whether the input web page contains information related to the research purpose based on the learning result of the classifier model, and classifies the input web page based on the determination result; a research execution unit that instructs a generation AI to extract information related to the research purpose from web pages that are classified by the web page classification unit as having information related to the research purpose among the input web pages; A database that stores information related to the research objective extracted by the generating AI; An information collection device comprising:

[0053] (Appendix 2) The investigation execution unit evaluates the information related to the investigation purpose extracted by the generation AI based on predetermined conditions, and selects highly credible information based on the evaluation results; 2. The information collection device according to claim 1, wherein the database stores the highly credible information selected by the investigation execution unit.

[0054] (Appendix 3) The information collection device of Appendix 2, wherein the investigation execution unit evaluates the credibility of information related to the investigation purpose based on at least one of the number of similar data, the number of contradictory data, and the last update date of the reference source as the predetermined conditions.

[0055] (Appendix 4) The information collection device of appendix 1 to 3, wherein the investigation execution unit refers to the labeling data, and if there is unlearned data that has not been used to learn the classifier model, instructs the learning unit to learn the classifier model.

[0056] (Appendix 5) The information collection device according to any one of claims 1 to 4, wherein the research execution unit repeats the extraction of information by the generation AI a predetermined number of times for web pages classified as containing information related to the research purpose.

[0057] (Appendix 6) The information collection device according to any one of claims 1 to 5, wherein the learning unit refers to the labeling data, extracts text from the arbitrary web page, converts it into a vector, and uses the converted vector and a label associated with the arbitrary web page to train the classifier model.

[0058] (Appendix 7) The information collection device of any one of appendices 1 to 6, wherein the web page classification unit extracts text from the input web page, converts the extracted text into a vector, and classifies the converted vector using the classifier model based on whether it is relevant to the research purpose.

[0059] (Appendix 8) The data is collected from the website and further includes crawling data for determining whether there are any differences in the web pages since the previous survey; The information collection device of appendix 1 to 7, wherein the research execution unit refers to the crawling data, and if there is a difference between the input web page and the previous research execution, instructs the generation AI to extract information related to the research purpose from the web page that has been classified by the web page classification unit as containing information related to the research purpose.

[0060] (Appendix 9) A step of training a classifier model to determine whether an input web page contains information related to the research purpose, based on labeling data that indicates, using a logical value, whether a given web page is related to the research purpose; classifying the input web page based on whether the input web page contains information relevant to the research objective based on the learning result of the classifier model; Instructing the generation AI to extract information related to the research purpose from web pages classified as containing information related to the research purpose among the input web pages; Storing the information related to the research objective extracted by the generation AI in a database; An information gathering method comprising:

[0061] (Appendix 10) A step of evaluating the information related to the research purpose extracted by the generation AI based on predetermined conditions and selecting highly credible information based on the evaluation results; 10. The information collection method of claim 9, further comprising the step of storing the selected highly credible information in the database.

[0062] (Appendix 11) 11. The information gathering method of claim 10, wherein the credibility of information related to the research purpose is evaluated based on at least one of the predetermined conditions: the number of similar data, the number of contradictory data, and the last update date of the reference source.

[0063] (Appendix 12) The information collection method of appendices 9 to 11, further comprising a step of referring to the labeling data and instructing training of the classifier model if there is untrained data that has not been used to train the classifier model.

[0064] (Appendix 13) The information collection method of claims 9 to 12, further comprising a step of repeating information extraction by the generation AI a predetermined number of times for web pages classified as containing information related to the research purpose.

[0065] (Appendix 14) The information gathering method of any one of claims 9 to 13, wherein the step of training the classifier model comprises: referring to the labeling data, extracting text from the arbitrary web page and converting it into a vector; and training the classifier model using the converted vector and a label associated with the arbitrary web page.

[0066] (Appendix 15) 15. The information gathering method of claim 9, wherein the step of classifying the input web page comprises extracting text from the input web page, converting the extracted text into vectors, and classifying the converted vectors using the classifier model based on whether they are relevant to the research purpose.

[0067] (Appendix 16) The information collection method described in Appendices 9 to 15, characterized in that the step of instructing the generation AI refers to crawling data, which is data collected from a website, to determine whether there are any differences in the web page since the last time the search was performed, and if there are any differences in the input web page since the last time the search was performed, instructs the generation AI to extract information related to the search purpose from web pages classified as containing information related to the search purpose.

[0068] (Appendix 19) The computer of the information collection device, a learning function that trains a classifier model to determine whether an input web page contains information related to the research objective, based on labeling data that indicates, using a logical value, whether a given web page is related to the research objective; a web page classification function that classifies the input web page based on whether the input web page contains information related to the research purpose, based on the learning result of the classifier model; a research execution function that instructs the generation AI to extract information related to the research purpose from web pages classified as containing information related to the research purpose among the input web pages; A storage function for storing information related to the research purpose extracted by the generating AI in a database; A program characterized by functioning as

[0069] (Appendix 20) The investigation execution function evaluates the information related to the investigation purpose extracted by the generation AI based on predetermined conditions, and selects highly credible information based on the evaluation results. The program of claim 19, wherein the storage function stores the highly credible information selected by the investigation execution function in the database.

[0070] (Appendix 21) The program of Appendix 20, wherein the investigation execution function evaluates the credibility of information related to the investigation purpose based on at least one of the predetermined conditions: the number of similar data, the number of contradictory data, and the last update date of the reference source.

[0071] (Appendix 22) The program of any one of appendices 19 to 21, wherein the investigation execution function refers to the labeling data, and if there is unlearned data that has not been used to learn the classifier model, instructs the learning function to learn the classifier model.

[0072] (Appendix 23) The program of appendix 19 to 22, wherein the research execution function repeats information extraction by the generation AI a predetermined number of times for web pages classified as containing information related to the research purpose.

[0073] (Appendix 24) The program of any one of appendices 19 to 23, wherein the learning function refers to the labeling data, extracts text from the arbitrary web page, converts it into a vector, and uses the converted vector and a label associated with the arbitrary web page to train the classifier model.

[0074] (Appendix 25) The program of any one of claims 19 to 24, wherein the web page classification function extracts text from the input web page, converts the extracted text into vectors, and classifies the converted vectors using the classifier model based on whether they are relevant to the research objective.

[0075] (Appendix 26) The program of appendix 19 to 25, wherein the research execution function refers to crawling data, which is data collected from a website, to determine whether there are any differences in the web page since the previous research, and if there are any differences in the input web page since the previous research was performed, instructs the generation AI to extract information related to the research purpose from web pages that have been classified by the web page classification function as containing information related to the research purpose. [Explanation of symbols]

[0076] 1,100 Information gathering device 10 servers / container 11 Scheduler 12 Processes 121, 106 Research Execution Department 13 modules 131, 103 Learning Department 132, 104 Web page classification section 20 databases / models 21, 107 Database 211, 101 labeling data 212 Homepage Data 213 Extracted Data 214 Survey Results Data 215 crawling data 22 Classifier Model 30, 105 Generation AI

Claims

1. a learning unit that learns a classifier model to determine whether an input web page contains information related to the research purpose, based on labeling data that indicates, using a logical value, whether a given web page is related to the research purpose; and a web page classification unit that determines whether the input web page contains information related to the research purpose based on the learning result of the classifier model, and classifies the input web page based on the determination result; a research execution unit that instructs a generation AI to extract information related to the research purpose from web pages that are classified by the web page classification unit as having information related to the research purpose among the input web pages; a database for storing information related to the research objective extracted by the generating AI; An information collection device comprising:

2. The investigation execution unit evaluates the information related to the investigation purpose extracted by the generation AI based on predetermined conditions, and selects highly credible information based on the evaluation results; 2. The information collection device according to claim 1, wherein the database stores the highly credible information selected by the investigation execution unit.

3. The information collection device described in claim 2, characterized in that the investigation execution unit evaluates the credibility of information related to the investigation purpose based on at least one of the specified conditions: the number of similar data, the number of contradictory data, and the last update date of the reference source.

4. 2. The information collection device according to claim 1, wherein the survey execution unit refers to the labeling data, and if there is unlearned data that has not been used to learn the classifier model, instructs the learning unit to learn the classifier model.

5. The information collection device according to claim 1, characterized in that the research execution unit repeats the extraction of information by the generation AI for web pages classified as containing information related to the research purpose a predetermined number of times.

6. 2. The information collection device according to claim 1, wherein the learning unit refers to the labeling data, extracts text from the arbitrary web page, converts the extracted text into a vector, and uses the converted vector and a label associated with the arbitrary web page to learn the classifier model.

7. 2. The information collection device according to claim 1, wherein the web page classification unit extracts text from the input web page, converts the extracted text into a vector, and classifies the converted vector using the classifier model based on whether it is relevant to the research purpose.

8. The data is collected from the website and further includes crawling data for determining whether there are any differences in the web pages since the previous survey; The information collection device described in claim 1, characterized in that the survey execution unit refers to the crawling data, and if there is a difference between the input web page and the previous survey, instructs the generation AI to extract information related to the survey purpose from web pages classified by the web page classification unit as containing information related to the survey purpose.

9. A step of training a classifier model to determine whether an input web page contains information related to the research purpose, based on labeling data that indicates, using a logical value, whether a given web page is related to the research purpose; classifying the input web page based on whether the input web page contains information relevant to the research objective based on the learning result of the classifier model; Instructing a generation AI to extract information related to the research purpose from web pages classified as containing information related to the research purpose among the input web pages; storing the information related to the research objective extracted by the generating AI in a database; An information gathering method comprising:

10. The computer of the information collection device, a learning function that trains a classifier model to determine whether an input web page contains information related to the research objective, based on labeling data that indicates, using a logical value, whether a given web page is related to the research objective; a web page classification function that classifies the input web page based on whether the input web page contains information related to the research purpose, based on the learning result of the classifier model; A research execution function that instructs the generation AI to extract information related to the research purpose from web pages classified as containing information related to the research purpose among the input web pages; A storage function for storing information related to the research objective extracted by the generating AI in a database; A program characterized by functioning as

Images