Evaluation support device, evaluation method, and program

Abstract

A propensity to classification of chemical substance harmfulness may be evaluated or predicted. A control unit provided in the evaluation support device outputs information for evaluating or predicting a tendency for classifying the harmfulness of the chemical substance, on the basis of the relevance of the plurality of documents.

Description

Technical Field

[0001] The present invention relates to an evaluation support device, an evaluation method and a program. Background Art

[0002] There are technologies for predicting the toxicity of chemical substances based on their structural features. For example, Patent Document 1 discloses an invention that vectorizes the structure of a chemical substance and calculates a toxicity prediction score using a learned classifier.

[0003] [Prior art literature]

[0004] [Patent Literature]

[0005] [Patent Document 1] International Publication No. 2018 / 049376 Summary of the invention

[0006] [Problems to be solved by the invention]

[0007] However, the standards for regulating chemical substances may change due to external factors such as social factors in addition to the toxicity of the chemical substances themselves. Therefore, it is difficult to evaluate or predict the toxicity of chemical substances based only on their structural characteristics.

[0008] The present invention can evaluate or predict the tendency of classification regarding the toxicity of chemical substances.

[0009]

Methods to solve the problem

[0010] An evaluation support device according to a first aspect of the present invention is an evaluation support device including a control unit configured to output information for evaluating or predicting a tendency of classification regarding toxicity of a chemical substance based on correlation between a plurality of documents.

[0011] According to the first aspect of the present invention, it is possible to evaluate or predict the tendency of classification regarding the toxicity of chemical substances.

[0012] A second aspect of the present invention is an evaluation support device according to the first aspect, wherein the control unit statistically processes information obtained by classifying the plurality of documents according to the properties of each chemical substance, thereby outputting information for evaluating or predicting the tendency of the classification regarding the toxicity of the chemical substance.

[0013] A third aspect of the present invention is the evaluation support device according to the second aspect, wherein the control unit classifies the plurality of documents based on whether or not the property of the chemical substance is described.

[0014] A fourth aspect of the present invention is the evaluation support device according to the third aspect, wherein the control unit performs statistical processing on information obtained by classifying the plurality of documents based on characteristic information of the chemical substances described in the documents.

[0015] A fifth aspect of the present invention is the evaluation support device according to the fourth aspect, wherein the control unit evaluates or predicts a tendency of the classification regarding the toxicity of the chemical substance based on the output information.

[0016] A sixth aspect of the present invention is the evaluation support device according to the fifth aspect, wherein the control unit evaluates or predicts the tendency of chemical substances having similar characteristic information based on characteristic information of the chemical substances recorded in the document.

[0017] The seventh aspect of the present invention is an evaluation support device according to the second to sixth aspects, wherein the properties of the chemical substance include at least one of the toxicity, bioaccumulation, difficulty in decomposition, geographical distribution, flammability and greenhouse effect of the chemical substance.

[0018] An eighth aspect of the present invention is the evaluation support device according to any one of the second to sixth aspects, wherein the properties of the chemical substance include information that reduces compliance with classification related to the toxicity of the chemical substance.

[0019] A ninth aspect of the present invention is the evaluation support device according to the eighth aspect, wherein the information for reducing the compliance includes at least one of a method for handling and a method for decomposing the chemical substance.

[0020] A tenth aspect of the present invention is the evaluation support device according to the second aspect, wherein the control unit classifies the plurality of documents based on a distributed representation in which the documents having high mutual relevance are arranged close to each other.

[0021] The 11th mode of the present invention is an evaluation support device according to the 10th mode, wherein the control unit assigns document information indicating whether the properties of the chemical substance are recorded to a part of the multiple documents, and classifies the documents that record the properties of the chemical substance based on the document information.

[0022] A twelfth aspect of the present invention is the evaluation support device according to any one of the second to eleventh aspects, wherein the control unit classifies the plurality of documents based on a reference relationship between the plurality of documents.

[0023] A thirteenth aspect of the present invention is the evaluation support device according to the twelfth aspect, wherein the document is an academic paper.

[0024] A fourteenth aspect of the present invention is the evaluation support device according to any one of the second to thirteenth aspects, wherein the control unit classifies the plurality of documents based on a result of natural language processing of the contents of the documents.

[0025] The evaluation method according to the fifteenth aspect of the present invention includes the step of evaluating or predicting the tendency of classification regarding the toxicity of a chemical substance based on the correlation between a plurality of documents.

[0026] A program according to a sixteenth aspect of the present invention causes a control unit included in an evaluation support device to execute the following steps to output information for evaluating or predicting a tendency of classification regarding toxicity of a chemical substance based on correlation between a plurality of documents. BRIEF DESCRIPTION OF THE DRAWINGS

[0027]

Figure 1

[0028]

Figure 2

[0029]

Figure 3

[0030]

Figure 4

[0031]

Figure 5

[0032]

Figure 6

[0033]

Figure 7

[0034] Hereinafter, each embodiment will be described with reference to the attached Figure 1 In addition, in this specification and drawings, components having substantially the same functional configuration are given the same reference numerals, and duplicate descriptions are omitted.

[0035] [Implementation Method]

[0036] This embodiment is an evaluation support device that outputs information for evaluating or predicting the tendency of classification regarding the harmfulness of chemical substances. The evaluation support device in this embodiment learns classifiers for each classification regarding the harmfulness of chemical substances based on the correlation of a plurality of collected documents, and classifies the documents to be investigated. In addition, the evaluation support device in this embodiment statistically processes information on the classification of documents according to the properties of each chemical substance, thereby outputting information for evaluating or predicting the tendency of classification regarding the harmfulness of chemical substances.

[0037] The classification of harmfulness of chemical substances is a collection of chemical substances classified based on the nature of the harm they cause to humans, organisms or the environment. An example of the classification of harmfulness of chemical substances is, for example, whether they comply with CMR (carcinogenic, mutagenic or reproductive toxic) substances, PBT (persistent, bioaccumulative and toxic) substances or vPvB (very persistent and very bioaccumulative) substances in the European REACH (Registration, Evaluation, Authorization and Restriction of Chemicals) rules. In addition, CMR substances are chemical substances designated as substances that have an impact on human health. PBT substances are chemical substances designated as substances that are difficult to decompose, bioaccumulate or toxic and have an impact on the environment. vPvB substances are chemical substances designated as substances that are extremely difficult to decompose and have high bioaccumulation.

[0038] <System Structure>

[0039] Figure 1 1 is a block diagram showing an example of the system configuration of the evaluation support device 10 in this embodiment. Figure 1 As shown, document data including label data and survey object data is input to the evaluation support device 10. The evaluation support device 10 converts each input document data into a document vector, and learns a classifier for each category of chemical substance toxicity based on the correlation of the document data. The evaluation support device 10 classifies the input survey object data into various categories, and outputs information for evaluating or predicting the tendency of the classification regarding the toxicity of chemical substances based on the statistical information of each category.

[0040] The document data in this embodiment is data representing a document related to a chemical substance. An example of the document data is paper data representing the content of an academic paper related to a chemical substance. The paper data can be collected using a paper database or the like. For example, SCOPUS (registered trademark) can be used as the paper database.

[0041] Another example of document data is a collection of patent publications on chemical substances. The collection of patent publications may be a collection of publications issued by patent offices of various countries, or a database containing publications issued by patent offices of various countries may be used.

[0042] The document data may not represent the entire document (for example, an academic paper or a patent specification, etc.). The document data may be data representing a part of the document or data representing an abstract.

[0043] The document data is provided with characteristic information of the chemical substance recorded in the document. One example of characteristic information is identification information for identifying the chemical substance. Another example of characteristic information is fingerprints of the compound, or information on functional groups or skeletons.

[0044] Identification information for identifying a chemical substance, such as a compound name, a name based on the IUPAC (International Union of Pure and Applied Chemistry) nomenclature, a symbol based on the SMILES notation, an InChI (International Compound Identifier) ​​code (Key), or a structural formula, etc. The identification information is not limited to these, and all information that can identify a chemical substance can be used.

[0045] The characteristic information of chemical substances recorded in the document sometimes contains a lot of noise such as symbol changes. Symbol changes indicate that different characteristic information is assigned to the same substance. Therefore, the characteristic information assigned in the document data can be cleared of symbol changes using a chemical substance database or the like. An example of a chemical substance database is the Japanese Chemical Substance Dictionary.

[0046] The labeled data is document data with document information. The survey object data is document data without document information. The document information is information indicating whether the properties of the chemical substance are recorded in the document data. The document information can be a true value of binary classification of whether the document data meets each property.

[0047] The relevance of a document is based on the relevance of the contents recorded in the document data. The relevance of a document may be based on the properties of a chemical substance, etc.

[0048] An example of the properties of a chemical substance is information indicating the type of harmfulness of the chemical substance. The types of harmfulness of the chemical substance include, for example, at least one of toxicity, bioaccumulation, difficulty in decomposition, geographical distribution, flammability, and greenhouse effect of the chemical substance. The types of harmfulness of the chemical substance are not limited to these, and may include other types.

[0049] Another example of the properties of chemical substances is information indicating the type of environmental response technology. Environmental response technology includes, for example, a method for treating or decomposing a chemical substance. It is believed that by properly treating or decomposing a chemical substance, the harmfulness of the chemical substance is reduced. Therefore, the information on environmental response technology is information that reduces compliance with the type of harmfulness of the chemical substance. The properties of the chemical substance are not limited to these, and may also include other properties.

[0050] The statistical information in this embodiment is the result of summarizing documents classified into identification information about chemical substances and types of harmfulness. The statistical information may be the rate of increase in the number of copies of document data, the proportion of various types, the time series thereof, etc. The statistical information may be a summary of bibliographic information further based on the document. The bibliographic information is, for example, the year of publication, the issuing agency or the author, etc. The statistical information may be a summary based on the density of the network of the agency or the author.

[0051] An example of information for evaluating or predicting the classification of harmfulness of chemical substances may be information showing the time series change of the number of documents classified into various categories regarding a certain chemical substance. By referring to such information, the types of harmfulness of the chemical substance that are currently being discussed can be understood.

[0052] Another example of information for evaluating or predicting the classification of chemical substances' harmfulness may be information showing a list of chemical substances with a large number of documents among documents classified into various categories. By referring to such information, chemical substances that are currently being discussed in a certain category of harmfulness can be understood.

[0053] <Hardware Configuration>

[0054] Figure 2 1 is a block diagram showing an example of the hardware configuration of the evaluation support device 10 in this embodiment. Figure 2 As shown, the evaluation support device 10 includes a processor 101, a memory 102, an auxiliary storage device 103, an operation device 104, a display device 105, a communication device 106, and a drive device 107. In addition, the hardware components of the evaluation support device 10 are connected to each other via a bus 108.

[0055] The processor 101 includes various computing devices such as a CPU (Central Processing Unit), etc. The processor 101 reads various programs installed in the auxiliary storage device 103 to the memory 102 and executes them.

[0056] The memory 102 has a main storage device such as a ROM (read only memory) and a RAM (random access memory). The processor 101 and the memory 102 form a so-called computer (hereinafter also referred to as a "control unit"), and the processor 101 executes various programs read from the memory 102, thereby realizing various functions of the computer.

[0057] The auxiliary storage device 103 stores various programs and various data used when the various programs are executed by the processor 101 .

[0058] The operation device 104 is an operation device for the user of the evaluation support device 10 to perform various operations. The display device 105 is a display device for displaying the processing results of various processes executed by the evaluation support device 10.

[0059] The communication device 106 is a communication device for communicating with an external device via a network (not shown).

[0060] The drive device 107 is a device for setting the storage medium 109. The storage medium 109 mentioned here includes a medium that stores information optically, electrically or magnetically, such as a CD-ROM, a floppy disk, a magneto-optical disk, etc. In addition, the storage medium 109 may include a semiconductor memory that stores information electrically, such as a ROM, a flash memory, etc.

[0061] In addition, various programs installed in the auxiliary storage device 103 are installed, for example, by providing the distributed storage medium 109 in the drive device 107, and reading various programs stored in the storage medium 109 through the drive device 107. Alternatively, various programs installed in the auxiliary storage device 103 may be installed by downloading them from the network through the communication device 106.

[0062] <Functional structure>

[0063] Figure 3 1 is a block diagram showing an example of the functional configuration of the evaluation support device in this embodiment. Figure 3 As shown, the evaluation support device 10 in this embodiment includes an input unit 11 , a conversion unit 12 , a learning unit 13 , a classifier storage unit 14 , an extraction unit 15 , and an output unit 16 .

[0064] The classifier storage unit 14 is as follows Figure 2 The input unit 11, the conversion unit 12, the learning unit 13, the extraction unit 15 and the output unit 16 are implemented by the memory 102 or the auxiliary storage device 103 shown in FIG. Figure 2 The processor 101 shown is implemented by executing various programs read from the memory 102 .

[0065] The input unit 11 receives input of a plurality of document data. The document data includes a plurality of annotation data and a plurality of survey object data.

[0066] The conversion unit 12 generates a document vector for each document data received by the input unit 11 .

[0067] The learning unit 13 learns classifiers for each category regarding harmfulness based on the document vector generated by the conversion unit 12 .

[0068] The classifier storage unit 14 stores the classifiers of various types learned by the learning unit 13 .

[0069] The extraction unit 15 classifies the survey target data received by the input unit 11 into various categories using the classifier stored in the classifier storage unit 14 .

[0070] The output unit 16 outputs information for evaluating or predicting the harmfulness of a chemical substance based on various types of statistical information.

[0071] <Flow of evaluation support methods>

[0072] Figure 4 This is a flowchart showing an example of the flow of the evaluation support method in this embodiment.

[0073] In step S1 , the input unit 11 receives input of a plurality of document data. The document data includes a plurality of annotation data and a plurality of survey object data. Next, the input unit 11 sends the received document data to the conversion unit 12 .

[0074] The number of labeled data can be greater than the number required for classifier training. For example, the number of labeled data for each class can be about 100 to 500. The number of survey target data is not limited, but can be about 10,000, for example.

[0075] In step S2 , the conversion unit 12 receives the document data from the input unit 11 . Next, the conversion unit 12 generates a document vector for each document data. Next, the conversion unit 12 sends the generated plurality of document vectors to the learning unit 13 .

[0076] The document vector in this embodiment is composed of at least one of a distributed representation in a reference space and a distributed representation in a document space. The distributed representation in the reference space is generated based on the reference relationship between document data. An example of a distributed representation in the reference space is LINE (Large Scale Information Network Embedding). Details about LINE are disclosed in the following reference 1.

[0077] [Reference 1] Tang, J., Qu, M., Wang, M., Zhang, M., Yan, J., and Mei, Q. "Line: Large-scale information embedding in networks," in 24th International Conference on World Wide Web, pp. 1067-1077, 2015.

[0078] In addition, LINE has a distributed representation based on first-order similarity and a distributed representation based on second-order similarity. In this embodiment, both the distributed representation based on first-order similarity and the distributed representation based on second-order similarity can be used, but the distributed representation based on second-order similarity is more suitable.

[0079] The distributed representation based on first-order similarity is a probability representation that indicates whether pairs of nodes are connected to each other. The distributed representation based on first-order similarity is represented by equation (1). Here, ν represents a node, and u is a low-dimensional vector representation of ν.

[0080]

Number 1

[0081]

[0082] The distributed representation based on second-order similarity is a probability representation that indicates whether pairs of nodes have a common connection with other nodes. The distributed representation based on second-order similarity is expressed by formula (2). Here, ν represents a node, u is a low-dimensional vector representation of ν, and u' is a vector representation representing the context.

[0083]

Number 2

[0084]

[0085] The distributed representation in the document space is generated based on the content of the document data. An example of the distributed representation in the document space is BERT (Bidirectional Encoder Representation by Transformer). The details of BERT are disclosed in the following reference 2.

[0086] [Reference 2] Beltagy, I., Lo, K., and Cohan, A., "SciBERT: Pre-trained Language Models for Scientific Text," arXiv preprint, arXiv:1903.10676, 2019.

[0087] The conversion unit 12 may generate the distributed representation in the reference space as a document vector, or may generate the distributed representation in the document space as a document vector. In addition, the conversion unit 12 may generate the distributed representation in the reference space and the distributed representation in the document space respectively, and use the vector combining them as the document vector.

[0088] In step S3, the learning unit 13 receives a plurality of document vectors corresponding to each document data from the conversion unit 12. Next, the learning unit 13 configures each document vector in a multidimensional space. The multidimensional space in this embodiment is configured such that document vectors with high mutual correlation are arranged close to each other. At this time, all document vectors including the labeled data and the survey object data are configured in the multidimensional space.

[0089] In step S4, the learning unit 13 learns classifiers according to various categories in the multidimensional space where the document vectors are configured. The classifier in this embodiment is an interface that divides the multidimensional space into two spaces. The learning unit 13 learns the following types of interfaces in the following manner: among the document vectors configured in the multidimensional space, the document vectors corresponding to the labeled data of the type to be learned are concentrated in one side of the space divided by the interface.

[0090] The interface in this embodiment is learned by logistic regression. Here, the learning method of the interface is not limited to logistic regression, and any learning method can be used as long as it is a classifier that can perform binary classification on a multidimensional space.

[0091] The interfaces in this embodiment are learned by categories. For example, when using categories including toxicity, bioaccumulation, difficulty in decomposition, and geographical distribution of chemical substances, the control unit learns the interface for toxicity, the interface for bioaccumulation, the interface for difficulty in decomposition, and the interface for geographical distribution, respectively.

[0092] Figure 5 This is a conceptual diagram showing an example of an interface. Figure 5 In the figure, white dots represent the configuration of the labeled data in the multidimensional space, and black dots represent the configuration of the survey object data in the multidimensional space. Figure 5 As shown, the interface is learned so that the labeled data is concentrated as much as possible in the space on one side divided by the interface.

[0093] In addition, the interface of each category is relearned every time the labeled data and the survey object data are added. As described later, the labeled data can be added by assigning the classification result to the survey object data. The learning unit 13 can relearn the interface after adding the labeled data.

[0094] In step S5, the extraction unit 15 reads the classifiers (interfaces) of each category stored in the classifier storage unit 14. Next, the extraction unit 15 classifies the survey object data using the interface of each category. The extraction unit 15 classifies each survey object data by determining whether it matches the category for each category.

[0095] The extraction unit 15 assigns the classification results of the various categories to the survey object data. Next, the extraction unit 15 sends the survey object data to which the classification results have been assigned to the output unit 16 .

[0096] The extraction unit 15 can use the survey object data assigned with the classification results as the label data to learn the interface of each category again. At this time, new survey object data is collected and its document vector is arranged in a multidimensional space. In this way, the document data of each category on harmfulness can be increased, and the harmfulness of chemical substances can be evaluated with higher accuracy.

[0097] In step S6, the output unit 16 receives the survey object data assigned with the classification result from the extraction unit 15. Next, the output unit 16 generates various types of statistical information based on the classification result. The statistical information in this embodiment is, for example, statistical information summarizing the number of copies based on the identification information of the chemical substance and the type of harmfulness. The statistical information in this embodiment can also be summarized based on the bibliographic information of the document such as the year of publication, the issuing agency or the author.

[0098] Next, the output unit 16 outputs information for evaluating or predicting the tendency of classification regarding the toxicity of chemical substances based on the statistical information of various categories. An example of the output information is information indicating the change in the number of documents classified into various categories regarding a certain chemical substance. Another example of the output information is information indicating a list of chemical substances with a large number of documents among the documents classified into various categories.

[0099] The output unit 16 can evaluate or predict the tendency of the classification of the toxicity of the chemical substance based on the information for evaluating or predicting the tendency of the classification of the toxicity of the chemical substance, and output the result. The evaluation result of the toxicity is, for example, a score calculated according to a specified rule for each type of toxicity. The prediction result of the toxicity is information indicating a type with a high possibility of being discussed in the future for a certain chemical substance, or information indicating a chemical substance with a high possibility of being discussed in the future for a certain type.

[0100] The output unit 16 can evaluate or predict the tendency of harmfulness of chemical substances with similar characteristic information about chemical substances recorded in the document data. For example, if the compound names and chemical formulas are similar, there is a high possibility that the tendency of classification about harmfulness is also similar. Therefore, by simultaneously evaluating or predicting chemical substances with similar characteristic information, there is a possibility of obtaining useful information.

[0101] Figure 6 This is an example of an output result showing the change in the number of documents classified into each category. Figure 6 As shown in the example of the output result, the total number of papers and the number of papers of each category in each publication year are outputted in a comparable manner for a certain chemical substance. Based on this output result, for example, the recent prevalence of discussion on toxicity, especially for a certain chemical substance, can be understood.

[0102] Figure 7 This is an example of an output result showing a list of chemical substances with a large number of documents among the documents classified into various categories. Figure 7 As shown in another example of the output result, a list of substances with a large number of papers is output in descending order in terms of harmfulness types (e.g., toxicity). Based on this output result, for example, chemical substances that have been particularly discussed recently regarding their toxicity can be understood.

[0103] <Evaluation Method Flow>

[0104] The user of the evaluation support device 10 in this embodiment can evaluate or predict the tendency of classification regarding the toxicity of chemical substances using the evaluation support device 10. Hereinafter, an evaluation method using the evaluation support device 10 will be described.

[0105] The user of the evaluation support device 10 inputs a plurality of document data into the evaluation support device 10. The evaluation support device 10 receives the input of the plurality of document data ( Figure 4 The evaluation support device 10 executes steps S2 to S5 of the evaluation support method based on the received input of the plurality of document data. Furthermore, the evaluation support device 10 outputs information ( Figure 4 Step S6).

[0106] The user of the evaluation support device 10 evaluates or predicts the tendency of the classification of the toxicity of the chemical substance based on the information output from the evaluation support device 10. For example, the user of the evaluation support device 10 can understand the chemical substances with popular toxicity discussions or the classification of toxicity with popular discussions about the specific chemical substances by referring to the classification statistical information about toxicity with respect to the specific chemical substances.

[0107] Summary

[0108] As described above, according to each embodiment of the present invention, the tendency of classification regarding the harmfulness of chemical substances can be evaluated or predicted. For example, the harmfulness of chemical substances, in addition to the toxicity of the chemical substances themselves, sometimes changes due to external reasons such as social reasons. Therefore, it is difficult to evaluate the harmfulness only from the characteristics of the structure of the chemical substance. The evaluation support device in this embodiment classifies multiple documents based on their association, and outputs information for evaluating or predicting the tendency of classification regarding the harmfulness of chemical substances based on statistical information using the properties of the chemical substances recorded therein. Therefore, according to the evaluation support device in this embodiment, the tendency of classification regarding the harmfulness of chemical substances can be evaluated or predicted.

[0109] In particular, the evaluation support device in this embodiment outputs information for evaluating or predicting the tendency of classification regarding the harmfulness of chemical substances based on statistical information on the properties of chemical substances recorded in documents using classification. With respect to a specific chemical substance, by referring to the statistical information on the classification regarding harmfulness, chemical substances with popular harmfulness discussions or classification regarding harmfulness of a specific chemical substance with popular harmfulness discussions are understood.

[0110] In addition, the evaluation support device in this embodiment classifies documents based on at least one of the toxicity, bioaccumulation, difficulty in decomposition, regional distribution, flammability, and greenhouse effect of the chemical substance. The harmfulness of chemical substances is discussed from various viewpoints, and thus by simultaneously evaluating the tendencies of various classifications, the harmfulness of chemical substances can be carefully evaluated.

[0111] Furthermore, the evaluation support device in this embodiment classifies academic papers based on citation relationships or natural language processing results. Although the harmfulness of chemical substances has been discussed from various viewpoints, it is not known whether it can be classified based on the citation relationship of academic papers. By classifying academic papers based on citation relationships or natural language processing results, it is expected that highly reliable information can be obtained for evaluating the harmfulness of chemical substances.

[0112] It will be appreciated that the above describes the embodiments, but various modifications in form and detail may be made without departing from the spirit and scope of the claims.

[0113] This application claims priority based on Japanese Patent Application No. 2022-86247 filed with the Japan Patent Office on May 26, 2022, the entire contents of which are incorporated herein by reference.

[0114] [Reference Signs]

[0115] 10 Evaluation support device

[0116] 11 Input section

[0117] 12 Conversion Department

[0118] 13 Learning Department

[0119] 14 Classifier storage unit

[0120] 15 Extraction Department

[0121] 16 Output section.

Claims

1. An evaluation support device comprising a control unit, The control unit outputs information for evaluating or predicting a tendency of classification regarding toxicity of a chemical substance based on correlation between a plurality of documents.

2. The evaluation support device according to claim 1, wherein the control unit By performing statistical processing on the information obtained by classifying the plurality of documents according to the properties of each chemical substance, Thus, information for evaluating or predicting the tendency of classification regarding the toxicity of chemical substances is output. 3 . The evaluation support device according to claim 2 , wherein the control unit classifies the plurality of documents based on whether or not the property of the chemical substance is described. 4 . The evaluation support device according to claim 3 , wherein the control unit performs statistical processing on information obtained by classifying the plurality of documents based on characteristic information of the chemical substances described in the documents. 5 . The evaluation support device according to claim 4 , wherein the control unit evaluates or predicts a tendency of the classification regarding the toxicity of the chemical substance based on the output information. 6 . The evaluation support device according to claim 5 , wherein the control unit evaluates or predicts the tendency of chemical substances having similar characteristic information based on characteristic information of the chemical substances described in the document.

7. The evaluation support device according to claim 2, wherein the properties of the chemical substance include at least one of toxicity, bioaccumulation, difficulty in decomposition, geographical distribution, flammability, and greenhouse effect of the chemical substance. 8 . The evaluation support device according to claim 2 , wherein the properties of the chemical substance include information that reduces compliance with classification regarding toxicity of the chemical substance. 9 . The evaluation support device according to claim 8 , wherein the information for reducing the compliance includes at least one of a method for handling and a method for decomposing the chemical substance. 10 . The evaluation support device according to claim 2 , wherein the control unit classifies the plurality of documents based on a distributed representation in which the documents having high mutual relevance are arranged close to each other.

11. The evaluation support device according to claim 10, wherein the control unit to provide document information indicating whether the properties of the chemical substance are described in some of the plurality of documents; The documents describing the properties of the chemical substances are classified based on the document information. 12 . The evaluation support device according to claim 2 , wherein the control unit classifies the plurality of documents based on a reference relationship between the plurality of documents.

13. The evaluation support device according to claim 12, wherein the document is an academic paper. 14 . The evaluation support device according to claim 2 , wherein the control unit classifies the plurality of documents based on a result of natural language processing of the contents of the documents.

15. An evaluation method comprising the steps of evaluating or predicting a tendency of classification regarding the toxicity of a chemical substance based on the correlation between a plurality of documents.

16. A program for causing a control unit of an evaluation support device to execute the following steps: Based on the correlation between the plurality of documents, information for evaluating or predicting the tendency of classification regarding the toxicity of the chemical substance is output.

Images