Information analysis system, information analysis method, and program

Abstract

This information analysis system comprises a co-occurrence matrix generation unit that generates, from text data set related to a keyword to be analyzed, a co-occurrence matrix which represents a frequency at which a pair of words appears together, a phrase generation unit that generates, on the basis of the text data set and in consideration of context before and after the words, a phrase obtained by expanding on the words, and an output unit that outputs a second co-occurrence network obtained by replacing, with the phrase obtained by expanding on the words, the words in a first co-occurrence network in which, on the basis of the co-occurrence matrix, the words are treated as nodes and the pair of words is connected via an edge.

Description

Information Analysis System, Information Analysis Method, and Program

[0001] The present invention relates to an information analysis system, an information analysis method, and a program.

[0002] There are information analysis systems that analyze various large amounts of text information. For example, there is a service that visualizes the relationship between a word and other words or presents the frequency of word occurrences by using a co-occurrence network that connects words as nodes and other words that appear in the same sentence as the word with edges (see, for example, Non-Patent Document 1).

[0003] Nomura Research Institute, "TRAINA Text Mining", [Searched on December 19, Reiwa 6], Internet, <URL:https: / / www.traina.ai / solution / textmining / >

[0004] However, in the conventional technology, when visualizing a large amount of text information by a co-occurrence network, since the words in the sentence are displayed as they are as nodes, there is a problem that it is difficult for an analyst to grasp the meaning of the context in which the word was used.

[0005] An embodiment of the present invention has been made in view of the above problems, and when visualizing a large amount of text information by a co-occurrence network, it makes it easier for an analyst to grasp the meaning of the sentence in which the word in the text was used.

[0006] [[ID=IS]] To solve the above problems, an information analysis system according to an embodiment of the present invention includes a co-occurrence matrix generation unit that generates a co-occurrence matrix representing the frequency of simultaneous appearance of word pairs from a text data group related to a keyword to be analyzed, a phrase generation unit that generates a phrase obtained by expanding the word in consideration of the context before and after the word based on the text data group, and an output unit that outputs a second co-occurrence network in which the words of a first co-occurrence network that connects the words as nodes and the word pairs as edges based on the co-occurrence matrix are replaced with the phrases obtained by expanding the words.

[0007] According to an embodiment of the present invention, when visualizing text data using a co-occurrence network, it becomes easier for the analyst to grasp the meaning of the sentences in which the words in the text were used.

[0008] This figure shows an example of the system configuration of the information analysis system according to this embodiment. This figure illustrates the co-occurrence matrix generation unit according to this embodiment. This figure illustrates the phrase generation processing unit according to this embodiment. This figure illustrates the processing of the output unit according to this embodiment. This flowchart shows an example of the information analysis process according to Example 1. This figure illustrates the information collection process according to Example 1. This figure illustrates the pre-processing according to Example 1. This figure illustrates the co-occurrence matrix generation process according to Example 1. This figure shows an example of a prompt according to Example 1. This figure shows an example of the system configuration of the information analysis system according to Example 2. This flowchart shows the flow of the information analysis process according to Example 2. This figure illustrates the clustering process according to Example 2. This figure shows an example of the computer hardware configuration.

[0009] Hereinafter, embodiments of the present invention (this embodiment) will be described with reference to the drawings. The embodiments described below are merely examples, and the embodiments to which the present invention is applied are not limited to the embodiments described below.

[0010] <Background and Challenges> For example, in the process of planning and formulating management strategies by companies, those in charge collect and analyze various types of information and formulate optimal strategies based on that information. However, there are limits to processing and analyzing large amounts of information manually, and currently, important information that would be useful for planning and formulating strategies is not being accurately analyzed.

[0011] To make it easier for people to analyze large amounts of diverse text information, technologies such as text mining have been developed. One example of this is the co-occurrence network. A co-occurrence network is a technology that visualizes the relationships between words as a network, or presents the frequency of word occurrences, by treating words as "nodes" from a large amount of text information and connecting them with "edges" that connect them to other words that appear in the same sentence. Furthermore, services that use this technology to enable people to accurately analyze information are known (see, for example, Non-Patent Document 1).

[0012] However, conventional technologies, when visualizing large amounts of text information using co-occurrence networks, display words in the text as nodes, making it difficult for analysts to grasp the context in which those words were used.

[0013] Therefore, the information analysis system 1 according to this embodiment has a system configuration, for example, as shown in Figure 1, in order to make it easier for the analyst to grasp the meaning of the sentences in which the words in the text were used when visualizing a large amount of text information using a co-occurrence network.

[0014] <System Configuration> Figure 1 shows an example of the system configuration of the information analysis system according to this embodiment. The information analysis system 1 is a system that analyzes and visualizes a large amount of text information. In the example in Figure 1, the information analysis system 1 includes an information analysis device 100 and LLM (Large Language Models) 110 that can be used from the information analysis device 100 via a communication network N. However, this is just an example. For example, the LLM 110 may be provided by an external service outside of the information analysis system 1. Also, the information analysis device 100 may be composed of multiple information processing devices.

[0015] (Information Analysis Device) The information analysis device 100 is an information processing device equipped with a computer configuration, or a system composed of multiple computers. The information analysis device 100 realizes, for example, the functional configurations shown in Figure 1 by executing a predetermined program. In the example in Figure 1, the information analysis device 100 has functional configurations such as an input unit 101, a collection unit 102, a preprocessing unit 103, a co-occurrence matrix generation unit 104, a phrase generation unit 105, an output unit 106, a communication unit 107, an information DB (Database) 108, and a text data storage unit 109. At least a part of the above functional configurations may be realized by hardware.

[0016] The input unit 101 performs input processing to receive keywords to be analyzed. For example, the input unit 101 receives keywords for the theme to be analyzed from the user.

[0017] The collection unit 102 performs a collection process to collect information related to the keyword entered in the input unit 101. For example, the collection unit 102 searches the internet or other sources based on the entered keyword, collects information related to the keyword by scraping or other means, and stores the collected information in the information database 108 or the like. The information database 108 is a database that stores the information collected by the collection unit 102. The information database 108 may be located outside the information analysis device 100.

[0018] The preprocessing unit 103 performs preprocessing by deleting unnecessary data such as figures and tables from the keyword-related information collected by the collection unit 102, extracting text data related to the keyword, and storing it in the text data storage unit 109. The text data storage unit 109 is a storage unit that stores the text data extracted by the preprocessing unit 103. The text data storage unit 109 may be located outside the information analysis device 100.

[0019] The co-occurrence matrix generation unit 104 performs a co-occurrence matrix generation process to generate a co-occurrence matrix representing the frequency of occurrence of word pairs from a group of text data related to the keyword to be analyzed.

[0020] Figure 2 is a diagram illustrating the co-occurrence matrix generation unit according to this embodiment. The co-occurrence matrix generation unit 104 calculates the frequency of occurrence (co-occurrence frequency) of a certain word (word A) and another word (word B) appearing simultaneously in the same sentence or text, and records the calculation result in a co-occurrence matrix 210 as shown in Figure 2.

[0021] Based on this co-occurrence matrix 210, for example, as shown in Figure 2, a co-occurrence network (hereinafter referred to as the first co-occurrence network 220) can be created that represents the relationships between words by treating each word as a node 221 and connecting the nodes 221 with edges 222. In the example in Figure 2, the thicker the edge 222 between nodes 221, the higher the frequency of occurrence.

[0022] The phrase generation unit 105 performs a phrase generation process that generates extended phrases for each word, taking into account the context before and after each word in the co-occurrence matrix 210, based on a group of text data related to the keyword.

[0023] For example, as shown in Figure 3, the phrase generation unit 105 inputs a set of sentences 310 containing words included in the co-occurrence matrix 210 (e.g., "clean energy") into the LLM 110, and generates a phrase 320 that expands on the word, taking into account the context before and after the word. In the example in Figure 3, the LLM 110 determines the context from the set of sentences 310 containing "clean energy" and generates the phrase 320 "widespread adoption of clean energy technology". This phrase makes it easy for the user to understand the context in which the word "clean energy" is used.

[0024] The output unit 106 performs output processing based on the co-occurrence matrix 210 to output a second co-occurrence network in which the words in the first co-occurrence network 220, which consists of words as nodes and word pairs connected by edges, are replaced with phrases that extend the words. For example, the output unit 106 replaces the "words" in the first co-occurrence network 220 created from the co-occurrence matrix 210 as shown in Figure 2 with "phrases" generated by the phrase generation unit 105 to create and output the second co-occurrence network 400 as shown in Figure 4.

[0025] In the second co-occurrence network 400, each node 401 contains not only the word itself, but also phrases that expand on the word by considering the context surrounding it. This makes it easier for the user to grasp the "meaning" of the sentence, such as the context in which the word was used.

[0026] (LLM) LLM110 is a large-scale language model built using a large amount of text data and deep learning technology. LLM110 performs various tasks such as reading, understanding, generating, translating, summarizing, and answering questions. Here, we will use existing LLM110 models such as ChatGPT, Gemini, tuzumi, or LLaMA without any modifications.

[0027] <Processing Flow> Next, the processing flow of the information analysis method according to this embodiment will be explained.

[0028] [Example 1] Figure 5 is a flowchart showing an example of information analysis processing according to Example 1. This process shows, for example, an example of information analysis processing performed by the information analysis system 1 described in Figures 1 to 4.

[0029] In step S501, the input unit 101 accepts the user's input of a keyword to be analyzed (for example, "climate change"). Note that the keyword is not limited to a single word, but may also be a phrase such as "impacts of climate change".

[0030] In step S502, the collection unit 102 collects information related to the entered keyword.

[0031] Figure 6 is a diagram illustrating the information gathering process according to Embodiment 1. For example, the collection unit 102 performs a web search based on the input keywords and scrapes the sites included in the search results 600 to collect literature data related to the keywords, which is then stored in the information DB 108.

[0032] In step S503, the preprocessing unit 103 extracts text data related to keywords from the collected information.

[0033] Figure 7 is a diagram illustrating the preprocessing according to Example 1. For example, in step S502, the preprocessing unit 103 removes figures, tables, headers, footers, annotations, etc., from the literature data 710 stored in the information DB 108, and extracts only the text data 711 and 712 of the main body of the literature data 710. The preprocessing unit 103 also stores the text data 720 extracted from the multiple literature data 710 in the text data storage unit 109. As a result, the text data storage unit 109 stores text data related to the keywords to be analyzed. Preferably, at this time, the preprocessing unit 103 unifies any inconsistencies in the notation of the text data 720.

[0034] In step S504, the co-occurrence matrix generation unit 104 generates a co-occurrence matrix from a group of text data related to the keyword to be analyzed, representing the frequency at which pairs of words appear together (co-occurrence frequency).

[0035] Figure 8 is a diagram illustrating the co-occurrence matrix generation process according to Embodiment 1. The co-occurrence matrix generation unit 104 generates a group of text data 800 by dividing the text data stored in the text data storage unit 109 into sentence-level text data, and generates pairs of words contained in each sentence, using each sentence as the unit. The co-occurrence matrix generation unit 104 also counts how many times each generated word pair appears in all sentences as a frequency, and generates a co-occurrence matrix 210, for example, as described in Figure 2, from the word pair and the application frequency. Note that the above-described method for generating a co-occurrence matrix is ​​just one example. The co-occurrence matrix generation unit 104 may also generate the co-occurrence matrix by applying a word weighting method such as TF-IDF (Term Frequency - Inverse Document Frequency), BM25, or the Jaccard coefficient. In this embodiment, the following explanation will be given assuming that the co-occurrence matrix is ​​generated for "nouns".

[0036] In step S505, the phrase generation unit 105 generates phrases that expand on words, taking into account the context surrounding the words in the co-occurrence matrix generated by the co-occurrence matrix generation unit 104. The phrase generation unit 105 extracts the sentences from which the words in the co-occurrence matrix 210 were obtained, for example, from the co-occurrence matrix 210 generated in step S504, such as the one shown in Figure 2. For example, the phrase generation unit 105 extracts the sentences from which the words were obtained by searching the text data used as input in step S504 for the words in the co-occurrence matrix. The phrase generation unit 105 also inputs the set of sentences obtained for each word in the co-occurrence matrix into the LLM 110 and phrases the words using prompts, for example, as illustrated in Figure 9.

[0037] Figure 9 shows an example of a prompt according to Embodiment 1. System prompt 901 describes, for example, the purpose of generating a phrase by extending a given word, and that the purpose is to determine the context and extend the word into a phrase. System prompt 901 also describes the processing flow, such as understanding the context of the words from the input sentences and considering expressions that take the context of the words into account. Furthermore, system prompt 901 describes the output format instructions and specific input and output examples. User prompt 902 allows the user to specify the target word and the sentences.

[0038] As a result, the phrase generation unit 105 generates phrases that expand each word of each node 221 of the first co-occurrence network 220, for example, as shown in Figure 2.

[0039] In step S506, the output unit 106 replaces the words in the co-occurrence network (first co-occurrence network) based on the co-occurrence matrix generated by the co-occurrence matrix generation unit 104 with phrases, and outputs a co-occurrence network (second co-occurrence network) with phrases as nodes. For example, in the first co-occurrence network 220 corresponding to the co-occurrence matrix 210 shown in Figure 2, the output unit 106 replaces the words in each node 221 with phrases generated by the phrase generation unit 105, and generates and outputs the second co-occurrence network 400 shown in Figure 4.

[0040] For example, in the first co-occurrence network 220 as shown in FIG. 2, it is difficult to grasp in what context the words of each node 221 are used. On the other hand, in the second co-occurrence network 400 as shown in FIG. 4, for each node 401, a phrase obtained by expanding the word is displayed in consideration of the context before and after the word. Therefore, it becomes easy to grasp in what context the words included in each node 401 are used.

[0041] [Example 2] In Example 2, an example of a suitable process when there are multiple contexts of sentences used for generating phrases in the phrase generation process executed by the phrase generation unit 105 will be described.

[0042] As described above, the phrase generation unit 105 inputs, for example, as shown in FIG. 3, a set of sentences 310 including the words of the co-occurrence matrix 210 to the LLM 110 to generate a phrase 320 obtained by expanding the word. At this time, when there are multiple different contexts in the set of sentences 310 including the words included in the co-occurrence matrix 210, it may be difficult for the LLM 110 to generate an appropriate phrase.

[0043] Therefore, the information analysis apparatus 100 according to Example 2 includes each functional configuration as shown in FIG. 10. As shown in FIG. 10, the information analysis apparatus 100 according to Example 2 includes a clustering unit 1001 in addition to each functional configuration included in the information analysis apparatus 100 described in FIG. 1.

[0044] The clustering unit 1001 is realized, for example, by a program executed by a computer included in the information analysis apparatus 100, and executes a clustering process for clustering a set of sentences including the words included in the co-occurrence matrix for each context. Note that, for example, a known clustering method such as the k-means method or the BERTopic method can be applied as the clustering method.

[0045] Also, in Example 2, the phrase generation unit 105 generates a phrase with expanded words using the set of sentences in the largest cluster among the set of sentences clustered by the clustering unit 1001. Regarding other functional configurations of the information analysis apparatus 100 according to Example 2, they may be the same as those of the information analysis apparatus 100 described in FIG. 1.

[0046] <Processing flow> FIG. 11 is a flowchart showing an example of information analysis processing according to Example 2. Among the processes shown in FIG. 11, the processes of steps S501 to S504 and S506 are the same as the information analysis processing according to Example 1 described in FIG. 5, so the description is omitted here.

[0047] In step S1101, the clustering unit 1001 clusters the context for each sentence in the set of sentences including the words of the co-occurrence matrix generated by the co-occurrence matrix generation unit 104.

[0048] FIG. 12 is a diagram for explaining the clustering process according to Example 2. The clustering unit 1001 extracts the set of sentences that are the source for obtaining the pairs of words in the co-occurrence matrix. In the example of FIG. 12, an example where the clustering unit 1001 extracts the set of sentences 1201 including "clean energy" is shown.

[0049] Also, the clustering unit 1001 performs context clustering for each sentence in the extracted set of sentences. For example, as shown in FIG. 12, the clustering unit 1001 clusters (groups) the set of sentences 1201 including "clean energy" into a plurality of clusters 1201a, 1202b, 1202c,... with different contexts.

[0050] In step S1102, the phrase generation unit 105 generates a phrase with expanded words using the set of sentences in the largest cluster among the clusters clustered by the clustering unit 1001. In the example of FIG. 12, the phrase generation unit 105 inputs the set of sentences in the cluster 1202a with the largest number of sentences among the plurality of clusters 1201a, 1202b, 1202c,... to the LLM 110, and phrases the words using, for example, the prompt exemplified in FIG. 9.

[0051] As shown in Figure 11, the phrase generation unit 105 can generate phrases with expanded words by excluding clusters 1202b, 1202c, ... which have different contexts from cluster 1202a, as shown in Figure 12. Therefore, according to Embodiment 2, even if the set of sentences containing the words in the co-occurrence matrix has multiple different contexts, the LLM 110 can easily generate appropriate phrases.

[0052] <Hardware Configuration> The information analysis device 100 according to this embodiment has a hardware configuration of a computer 1300 as shown in Figure 13. The information analysis device 100 may be implemented by multiple computers 1300. Furthermore, the computer is not limited to a physical machine, but may be a virtual machine on the cloud, for example.

[0053] Figure 13 shows an example of a computer hardware configuration. In the example in Figure 13, the computer 1300 includes a drive device 1301, an auxiliary storage device 1302, a memory device 1303, a CPU 1304, an interface device 1305, a display device 1306, an input device 1307, and an output device 1308, all of which are interconnected by bus B. The computer 1300 may also include other processors such as a GPU (Graphics Processing Unit).

[0054] The program that enables processing on the computer 1300 is provided on a recording medium 1311, such as a CD-ROM or memory card. When the recording medium 1311 containing the program is set in the drive device 1301, the program is installed from the recording medium 1311 to the auxiliary storage device 1302 via the drive device 1301. However, the program does not necessarily have to be installed from the recording medium 1311; it may also be downloaded from another computer via a network. The auxiliary storage device 1302 stores the installed program as well as necessary files and data.

[0055] The memory device 1303 reads and stores a program from the auxiliary storage device 1302 when a program startup command is received. The CPU 1304 implements functions related to the information analysis device 100, for example, according to the program stored in the memory device 1303. The interface device 1305 is used as an interface for connecting to a communication network, etc. The display device 1306 displays a GUI (Graphical User Interface) etc., based on a program. The input device 1307 consists of a keyboard, mouse, buttons, and / or touch panel etc., and is used to input various operation commands. The output device 1308 outputs the calculation results.

[0056] The CPU 1304 may also be other processors, such as a DSP (Digital Signal Processor), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array).

[0057] <Effects of the Embodiment> According to this embodiment, when visualizing text data using a co-occurrence network, it becomes easier for the analyst to grasp the meaning of the sentences in which the words in the text were used. For example, when formulating and developing business strategies, by using the co-occurrence network output by the information analysis device 100 according to this embodiment, it becomes easier for the analyst to grasp the meaning of the sentences in which the words were used, so that business strategies can be formulated and developed based on more accurate analysis results.

[0058] <Summary of Embodiments> This specification discloses at least the following information analysis systems, information analysis methods, programs, and information analysis devices. (Section 1) An information analysis system comprising: a co-occurrence matrix generation unit that generates a co-occurrence matrix representing the frequency of simultaneous occurrence of word pairs from a group of text data related to a keyword to be analyzed; a phrase generation unit that generates a phrase by extending the word, taking into account the context before and after the word, based on the group of text data; and an output unit that outputs a second co-occurrence network in which the words in a first co-occurrence network, in which the words are nodes and the word pairs are connected by edges, are replaced with the phrases by extending the word, based on the co-occurrence matrix. (Section 2) The information analysis system according to Section 1, further comprising: a clustering unit that clusters a set of sentences containing the word from the group of text data according to context, wherein the phrase generation unit generates a phrase by extending the word using the set of sentences of the largest cluster from the clustered set of sentences. (Clause 3) An information analysis method in which a computer performs the following steps: a co-occurrence matrix generation process that generates a co-occurrence matrix representing the frequency of simultaneous occurrence of word pairs from a group of text data related to the keyword to be analyzed; a phrase generation process that generates phrases by extending the word, taking into account the context before and after the word, using the group of text data; and an output process that outputs a second co-occurrence network in which the words in a first co-occurrence network, in which the words are nodes and the word pairs are connected by edges, are replaced with the phrases that extend the word, based on the co-occurrence matrix. (Clause 4) A program, or a storage medium storing a program, that causes a computer to execute: a co-occurrence matrix generation process that generates a co-occurrence matrix representing the frequency of simultaneous occurrence of word pairs from a group of text data related to the keyword to be analyzed; a phrase generation process that generates phrases by extending the words, taking into account the context before and after the words, using the group of text data; and an output process that outputs a second co-occurrence network in which the words in a first co-occurrence network, in which the words are nodes and the word pairs are connected by edges, are replaced with the phrases that extend the words, based on the co-occurrence matrix.(Clause 5) An information analysis device comprising: a co-occurrence matrix generation unit that generates a co-occurrence matrix representing the frequency of simultaneous occurrence of word pairs from a group of text data related to a keyword to be analyzed; a phrase generation unit that generates phrases by extending the words, taking into account the context before and after the words, based on the group of text data; and an output unit that outputs a second co-occurrence network based on the co-occurrence matrix, in which the words in a first co-occurrence network, in which the words are nodes and the word pairs are connected by edges, are replaced with the phrases that extend the words.

[0059] Although this embodiment has been described above, the present invention is not limited to this specific embodiment, and various modifications and changes are possible within the scope of the gist of the invention as described in the claims.

[0060] 1. Information Analysis System 100. Information Analysis Device 101. Input Unit 102. Data Collection Unit 103. Preprocessing Unit 104. Co-occurrence Matrix Generation Unit 105. Phrase Generation Unit 106. Output Unit 107. Communication Unit 108. Information Database 109. Text Data Storage Unit 110. LLM 1300. Computer

Claims

1. An information analysis system comprising: a co-occurrence matrix generation unit that generates a co-occurrence matrix representing the frequency of simultaneous occurrence of word pairs from a group of text data related to a keyword to be analyzed; a phrase generation unit that generates phrases by extending the words, taking into account the context before and after the words, based on the group of text data; and an output unit that outputs a second co-occurrence network in which the words in a first co-occurrence network, in which the words are nodes and the word pairs are connected by edges, are replaced with the phrases that extend the words, based on the co-occurrence matrix.

2. The information analysis system according to claim 1, comprising: a clustering unit that clusters a set of sentences containing the word from the text data group according to context; and a phrase generation unit that generates a phrase extending the word using the set of sentences with the largest cluster from the clustered set of sentences.

3. An information analysis method in which a computer performs the following steps: a co-occurrence matrix generation process that generates a co-occurrence matrix representing the frequency of simultaneous occurrence of word pairs from a set of text data related to the keyword to be analyzed; a phrase generation process that generates phrases by extending the words, taking into account the context before and after the words, using the set of text data; and an output process that outputs a second co-occurrence network in which the words in a first co-occurrence network, in which the words are nodes and the word pairs are connected by edges, are replaced with the phrases that extend the words, based on the co-occurrence matrix.

4. A program that causes a computer to execute the following: a co-occurrence matrix generation process that generates a co-occurrence matrix representing the frequency of simultaneous occurrence of word pairs from a set of text data related to the keyword to be analyzed; a phrase generation process that generates phrases by extending the words, taking into account the context before and after the words, using the set of text data; and an output process that outputs a second co-occurrence network in which the words in a first co-occurrence network, where the words are nodes and the word pairs are connected by edges, are replaced with the phrases that extend the words, based on the co-occurrence matrix.

Images