Information processing device, information processing method, and program

Abstract

The present invention provides an information processing device comprising: a summarizing unit that generates summary text from original text, which is text that serves as the source of target information to be created; an encoding unit that generates a source feature, which is a feature of the summary text; and a retrieval unit that extracts other text related to the target information by referencing a database storing a plurality of features obtained from a plurality of other texts other than the original text and comparing the plurality of features with the source feature.

Description

Information processing device, information processing method, and program

[0001] The present invention relates to a technique for collecting information necessary for creating a roadmap.

[0002] In many cases, organizations, such as companies, create roadmaps that list the necessary steps to achieve their goals in chronological order. Roadmaps allow organizations to share their understanding of the path to the goal both within the organization and externally.

[0003] In general, in order to create a roadmap, a creator of a roadmap needs to obtain a wide range of information on external trends related to the roadmap that he or she is trying to create.

[0004] Tomas Mikolov, et al, Distributed Representations of Words and Phrases and their Compositionality, Oct 2013, URL; https: / / arxiv.org / pdf / 1301.3781v3.pdfYang Liu, "Fine-tune BERT for Extractive Summarization," Sep. 2019, URL: [1903.10318] Fine-tune BERT for Extractive Summarization (arxiv.org)Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov, Luke Zettlemoyer, "BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension," Oct. 2019, URL: [1910.13461] BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension (arxiv.org)

[0005] However, obtaining a wide range of related information requires a lot of effort, time, and other costs.

[0006] In particular, when creating a roadmap, in order to make future plans, the creator should refer to roadmaps from organizations other than the one they belong to. However, if the industry, country, or specialty of the other organization is different from that of the creator's organization, it is difficult to understand the roadmap of that organization.

[0007] In particular, when it comes to situations overseas, it is costly to comprehensively grasp all the relevant information. On the other hand, when creating a roadmap, the information related to the roadmap needs to be limited to the information that the creator considers important.

[0008] The above-mentioned problem is not limited to roadmap creation, but can occur in any field in which, when creating certain information, information related to that information must be obtained.

[0009] The present invention has been made in view of the above points, and has an object to provide a technique that makes it possible to obtain information related to information to be created at low cost.

[0010] According to the disclosed technology, an information processing device is provided that includes: a summarization unit that generates a summary sentence from an original sentence, which is a sentence that is the source of the target information to be created; an encoding unit that generates original features that are features of the summary sentence; and a search unit that refers to a database that stores multiple features obtained from multiple other sentences other than the original sentence, and compares the multiple features with the original features to extract other sentences related to the target information.

[0011] The disclosed technology provides a technology that makes it possible to obtain information related to information to be created at low cost.

[0012] FIG. 1 is a diagram illustrating an image of creating a roadmap. FIG. 2 is a diagram illustrating an example of the functional configuration of the information processing device 100. FIG. 3 is a flowchart for explaining the operation of the information processing device 100. FIG. 4 is a flowchart for explaining the operation of the information processing device 100. FIG. 5 is a diagram for explaining collection of information on other companies, etc. FIG. 6 is a diagram illustrating an example of the hardware configuration of the information processing device 100.

[0013] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The embodiment described below is merely an example, and the embodiment to which the present invention is applied is not limited to the following embodiment.

[0014] In the following, an example will be described in which the information to be created is a roadmap (which may also be called a scenario); however, the technology according to the present invention is not limited to cases in which the information to be created is a roadmap, and can be applied to the acquisition of related information when creating information in various fields. Note that the term "text" used below includes one or more sentences and represents a certain topic. Note that the term "text" used below may be replaced with "text."

[0015] In the following, the problems will be described in more detail first, and then the technology according to the present embodiment will be described. Note that the contents of the description of the problems below are not publicly known.

[0016] (Challenges) As mentioned above, when creating a roadmap for a company or other organization, it is first necessary to create a document that describes the goals or vision, and to obtain information related to the roadmap (for example, trends in the industry related to the company creating the roadmap). However, obtaining such information requires a lot of effort. Figure 1 shows an example of how a roadmap is created.

[0017] Research and development plans and future forecasts for technology fields related to the roadmap you are trying to create are often published by government agencies, etc. However, it is difficult to collect and extract a large amount of information while determining its relevance to the roadmap you are trying to create.

[0018] Furthermore, with the promotion of evidence-based policy making (EBPM), many government documents, such as roadmaps, are being made public. However, this public information is often very long and it is difficult to grasp the key points, such as the chronological flow of events.

[0019] Furthermore, for fields that are different from the field of the roadmap being created, specialized knowledge about the different field is required, making it difficult to obtain and understand correct information about the different field.

[0020] In order to obtain information related to the roadmap to be created, it is conceivable to vectorize all of the sentences (e.g., vision or goals) about the roadmap to be created using, for example, the technology disclosed in Non-Patent Document 1, and also vectorize sentences that are candidates for related information, and then determine the similarity between them. However, this method increases the dimension of the vector too much, resulting in a huge amount of data, making it an unrealistic method. Furthermore, even if the sentences, such as the vision of the roadmap to be created, and the sentences to be compared are on the same topic, it is difficult to compare the ideas of the two.

[0021] Furthermore, when creating a roadmap and planning future plans, it is necessary to understand the relationship between the roadmap to be created and other roadmaps, but extracting other roadmaps to use as reference requires a lot of cost.

[0022] The configuration and operation of the device for solving the above problems will be described in detail below.

[0023] (Device Configuration) Fig. 2 shows an example of the functional configuration of the information processing device 100 in this embodiment. As shown in Fig. 2, the information processing device 100 has an input unit 110, a summarizing unit 120, an encoding unit 130, a search unit 140, an output unit 150, and a storage unit 160. The storage unit 160 may also be called a database. The storage unit 160 may also be provided outside the information processing device 100. The information processing device 100 may also be called a roadmap creation support system.

[0024] (Device Operation) The operation of the information processing device 100 having the above configuration will be described below with reference to the flowcharts of FIGS.

[0025] <S101> As a preliminary step, the user creates a document expressing the vision (or goal) of the roadmap to be created. The user may also create a draft of the roadmap (for example, a draft document describing the development plan of a certain product in chronological order) or a document predicting the future of the roadmap.

[0026] The above various documents created in advance preparation are collectively referred to as “original roadmap documents.” Note that the “original roadmap documents” are not limited to the above, and for example, documents on risk requirements, documents on risk items, etc. may also be used as the “original roadmap documents.”

[0027] In S101, one or more original roadmap sentences are input to the information processing device 100 by the input unit 110. For example, two original roadmap sentences, a sentence expressing a vision (or goal) and a sentence representing a draft roadmap, are input to the information processing device 100. The input original roadmap sentences are stored in the storage unit 160.

[0028] <S102> In S102, the summarizing unit 120 reads out the original roadmap sentence from the storage unit 160, summarizes the original roadmap sentence, and generates a summary sentence. If multiple original roadmap sentences are input, each original roadmap sentence is summarized.

[0029] The summarizing unit 120 may also generate a summary based on the original roadmap text, taking into account time-series elements, by generating a summary based on inference. BERT, for example, can be used to generate this summary.

[0030] In this embodiment, abstract summarization technology using NLP (Natural Language Processing) is used for summarization processing by the summarization unit 120. Specifically, for example, BERT or GPT can be used (Non-Patent Documents 2 and 3). However, the summarization method is not limited to the abstract summarization technology.

[0031] Summarization can be broadly divided into extraction-based summarization and abstraction-based summarization. Extraction-based summarization is a method of shortening a text by extracting sentences that are considered important from the entire text, while abstraction-based summarization is a method of generating a short summary (new sentence) that captures the meaning of the entire text.

[0032] For example, let's take the following sentence, which expresses the vision (goal) for the roadmap: (https: / / www.meti.go.jp / press / 2022 / 03 / 20230330005 / 20230330005.html).

[0033] "This roadmap, based on scientific evidence, outlines steady low-carbon efforts, such as energy conservation, efficiency improvement, and fuel conversion, as well as future innovative technologies, with reference to domestic policies and international scenarios, along with the background and timeline, toward achieving carbon neutrality in Japan by 2050. We expect companies to refer to this roadmap when considering climate change measures using transition finance. We also believe that it will help financial institutions and other organizations determine whether a company's strategy and efforts toward decarbonization qualify as transition finance when raising funds." When the above sentence is summarized using abstraction summarization, for example, the following sentence is generated: "This roadmap aims to enable companies to achieve carbon neutrality by utilizing transition finance." Similarly, when the above sentence is summarized using extraction summarization, the following sentence is generated: "This roadmap outlines future innovative technologies aimed at achieving carbon neutrality."

[0034] Furthermore, if we use inference to generate sentences from the above sentence, we can generate a sentence such as, "We aim to achieve carbon neutrality through transition finance by 2050."

[0035] Each summary sentence generated in S102 is stored in the storage unit 160.

[0036] <S103> In S103, the encoding unit 130 encodes and vectorizes each of the one or more summary sentences generated in S102. That is, for example, if summary sentence A (abstract summary) and summary sentence B (summary including time-series information generated by inference) are generated in S102, the encoding unit 130 generates a vector for summary sentence A and a vector for summary sentence B. By vectorizing the summary sentences, it becomes possible to handle document features with a small amount of information. Note that a "vector" is an example of a feature. Information other than a "vector" may also be used as a feature.

[0037] Furthermore, the encoding unit 130 classifies and labels each of the one or more summary sentences generated in S102. Examples of classes used in classification include a class related to important related technologies, a class related to specialized fields, a class related to industries, a class related to organizations, and a class related to the type of sentence (e.g., future prediction, economic analysis, etc.). Note that multiple labels related to multiple classes may be assigned to one summary sentence. For example, summary sentence A may be assigned two labels, such as "specialty field = automobiles, type of sentence = future prediction."

[0038] In this embodiment, classification is performed, but an embodiment may be adopted in which classification is not performed.

[0039] The vectors and labels generated in S103 are stored in the storage unit 160 in association with the corresponding summary sentences.

[0040] <<Regarding processing of text other than the original roadmap text>> Before, after, or in parallel with the processing of S101 to S103 on the original roadmap text, the information processing device 100 performs the same processing as S101 to S103 on text other than the original roadmap text.

[0041] The flow of this process is shown in Figure 4. In Figure 4, the processing content in S201 to S203 is the same as the processing content in S101 to S103, except that the sentence to be processed is different. In other words, summarization, vectorization, and class classification are performed using NLP. In S204, the processing results in S201 to S203 (summary sentence, vector, label) are stored in the storage unit 160 (database). The information stored in the storage unit 160 may be the summary sentence, vector, and label, or in addition, the sentence before summarization may also be stored.

[0042] Sentences other than the original roadmap sentences that are the subject of processing in S201 to S204 will be called “roadmap-related candidate sentences.” For example, if the original roadmap sentence is a sentence about organization A (e.g., a certain company), the roadmap-related candidate sentences will be sentences describing visions (goals), roadmap sentences, sentences about future forecasts, sentences about risk requirements, sentences about risk items, etc. that have been published by organizations other than organization A (such as governments, companies, and industry associations in various countries).

[0043] The information processing device 100 acquires (inputs) a plurality of roadmap-related candidate sentences from, for example, websites of various countries and organizations, and executes the processes of S201 to S204 for each of the acquired plurality of roadmap-related candidate sentences.

[0044] An image of how roadmap-related candidate sentences are acquired is shown in Figure 5. In the example of Figure 5, there are organizations A to C that are engaged in development in a field (e.g., automobiles) related to the original roadmap sentence, and the image of how roadmap-related candidate sentences are acquired for each organization is shown.

[0045] 3, the search unit 140 calculates the similarity between a vector obtained from the original roadmap sentence (referred to as the original vector) and each of multiple vectors obtained from multiple roadmap-related candidate sentences (referred to as related candidate vectors) by referring to the storage unit 160, and extracts related candidate vectors having a high similarity to the original vector. The related candidate vector having a high similarity to the original vector may be the related candidate vector having the highest similarity to the original vector among the multiple related candidate vectors.

[0046] Alternatively, a search may be performed using a label. For example, if the label of the original vector is A, the search unit 140 may calculate the similarity only for a plurality of related candidate vectors labeled A among all related candidate vectors, and extract related candidate vectors having a high similarity to the original vector from the plurality of related candidate vectors labeled A.

[0047] If there are multiple original vectors, the search unit 140 performs the above process for each original vector.

[0048] The search unit 140 acquires summary sentences (referred to as related summary sentences) corresponding to the extracted related candidate vectors from the storage unit 140, and passes the related summary sentences to the output unit 150. The output unit 150 outputs the related summary sentences, for example, by displaying them on a display. In addition to or instead of this process, the search unit 140 may acquire roadmap-related candidate sentences corresponding to the extracted related candidate vectors from the storage unit 140, and pass the roadmap-related candidate sentences to the output unit 150, and the output unit 150 may output the roadmap-related candidate sentences.

[0049] The output unit 150 may also access a website on which the original text of the related summary text is published, and output (display) the original text together with the related summary text.

[0050] The above process allows users to easily obtain information related to the roadmap they are creating (such as the visions and roadmaps of other organizations). This allows users to appropriately revise and add to the vision statement or the roadmap draft. For example, users can add related industry trends to the roadmap draft.

[0051] (Hardware Configuration Example) The information processing device 100 described in this embodiment can be realized, for example, by causing a computer to execute a program. This computer may be a physical computer or a virtual machine on the cloud.

[0052] That is, the device can be realized by executing a program corresponding to the processing performed by the device using hardware resources such as a CPU and memory built into a computer. The program can be recorded on a computer-readable recording medium (such as a portable memory) and stored or distributed. The program can also be provided via a network such as the Internet or email.

[0053] Fig. 6 is a diagram showing an example of the hardware configuration of the computer. The computer in Fig. 6 includes a drive device 1000, an auxiliary storage device 1002, a memory device 1003, a CPU 1004, an interface device 1005, a display device 1006, an input device 1007, an output device 1008, and the like, all of which are interconnected by a bus BS. The computer may further include a GPU.

[0054] The program that realizes the processing on the computer is provided by a recording medium 1001, such as a CD-ROM or a memory card. When the recording medium 1001 storing the program is set in the drive device 1000, the program is installed from the recording medium 1001 to the auxiliary storage device 1002 via the drive device 1000. However, the program does not necessarily have to be installed from the recording medium 1001, but may be downloaded from another computer via a network. The auxiliary storage device 1002 stores the installed program as well as necessary files, data, etc.

[0055] The memory device 1003 reads and stores a program from the auxiliary storage device 1002 when an instruction to start the program is received. The CPU 1004 realizes functions related to the information processing device 100 in accordance with the program stored in the memory device 1003. The interface device 1005 is used as an interface for connecting to a network, etc. The display device 1006 displays a GUI (Graphical User Interface) or the like according to the program. The input device 1007 is composed of a keyboard, mouse, buttons, a touch panel, etc., and is used to input various operation instructions. The output device 1008 outputs the results of calculations.

[0056] (Effects of the Embodiment) As described above, the technology described in the present embodiment makes it possible to obtain information related to information to be created (for example, a roadmap) at low cost.

[0057] The following additional notes are provided regarding the above-described embodiments.

[0058] <Additional Notes> (Additional Item 1) An information processing device including a memory and at least one processor connected to the memory, wherein the processor generates a summary from an original sentence that is a source sentence of target information to be created, generates original features that are features of the summary sentence, and extracts other sentences related to the target information by referring to a database in which a plurality of features obtained from a plurality of other sentences other than the original sentence are stored and comparing the plurality of features with the original features. (Additional Item 2) The information processing device according to Additional Item 1, wherein the processor extracts other sentences corresponding to a feature of the plurality of feature quantities that has the highest similarity to the original feature. (Additional Item 3) The information processing device according to Additional Item 1 or 2, wherein the processor performs class classification on the summary sentence and assigns a label to the summary sentence. (Additional Item 4) The information processing device according to Additional Item 3, wherein the processor extracts other sentences related to the target information from a plurality of other sentences to which the same label as the label is assigned. (Supplementary Item 5) The information processing device of any one of Supplementary Items 1 to 4, wherein the target information is a roadmap. (Supplementary Item 6) An information processing method executed by an information processing device, comprising: a summarizing step of generating a summary sentence from an original sentence that is a sentence that is the source of the target information to be created; an encoding step of generating original features that are features of the summary sentence; and a retrieval step of referencing a database that stores a plurality of features obtained from a plurality of other sentences other than the original sentence, and comparing the plurality of features with the original features to extract other sentences related to the target information. (Supplementary Item 7) A non-transitory storage medium that stores a program for causing a computer to function as each unit in the information processing device of any one of Supplementary Items 1 to 5.

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

[0060] 100 Information processing device 110 Input unit 120 Summarizing unit 130 Encoding unit 140 Search unit 150 Output unit 160 Storage unit 1000 Drive device 1001 Recording medium 1002 Auxiliary storage device 1003 Memory device 1004 CPU 1005 Interface device 1006 Display device 1007 Input device 1008 Output device

Claims

1. An information processing device comprising: a summarization unit that generates a summary sentence from an original sentence, which is the sentence that is the source of target information to be created; an encoding unit that generates original features that are features of the summary sentence; and a search unit that references a database in which a plurality of features obtained from a plurality of other sentences other than the original sentence are stored, and compares the plurality of features with the original features to extract other sentences related to the target information.

2. The information processing device according to claim 1, wherein the search unit extracts other sentences corresponding to a feature quantity that has the highest similarity to the original feature quantity among the plurality of feature quantities.

3. The information processing device according to claim 1, wherein the encoding unit performs class classification on the summary text and assigns a label to the summary text.

4. The information processing device according to claim 3, wherein the search unit extracts other sentences related to the target information from a plurality of other sentences to which the same label as the label is assigned.

5. The information processing device according to claim 1, wherein the target information is a roadmap.

6. An information processing method executed by an information processing device, comprising: a summarizing step of generating a summary sentence from an original sentence, which is the sentence that is the source of target information to be created; an encoding step of generating original features that are features of the summary sentence; and a retrieval step of extracting other sentences related to the target information by referring to a database in which a plurality of features obtained from a plurality of sentences other than the original sentence are stored and comparing the plurality of features with the original features.

7. A program for causing a computer to function as each unit in an information processing device according to any one of claims 1 to 5.

Images