Document display support system, document display support method, and program for executing the method

Abstract

To estimate important words and highlight display them in a document of a specific field.SOLUTION: A document display support system includes: a database in which selection object words and non-selection object words are registered; a learned word selection model which has been subjected to mechanical learning to estimate whether a certain word is a selection object word or not; a text pre-processing section for cutting out each word from a received display object document; a word classification section for classifying each of the words into one of a selection object word, a non-selection object word, and an indefinite word, on the basis of the database; a text post-processing section for imparting a prescribed attribute to a predetermined word in the display object document and generating output data; and an output section for outputting the output data. The word selection model, when estimating a label indicating that the indefinite word classified by the word classification section is a selection object word, classifies the indefinite word into a selection object word, and the text post-processing section imparts the prescribed attribute to the classified selection object word.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a document display support technology, and more particularly to a document display support system and method for visually distinguishing and displaying specific words in a document in a specific field, and a program for executing said method. [Background technology]

[0002] Currently, a huge number of documents are stored in databases around the world. Usually, a user needs to read a document and understand its contents in order to determine whether a document is relevant to a particular field and is important. For example, in the field of cancer genome medicine, medical professionals such as doctors and researchers annotate gene mutations obtained by information analysis while referring to knowledge databases and integrated databases that accumulate information on gene mutations, drugs, clinical trials, etc., in order to create reports, for example. Among these, information related to gene mutations and clinical significance is almost always published in the form of papers, and medical professionals are required to read the difficult text of the papers and correctly understand it.

[0003] However, the number of papers that need to be referred to is enormous, the contents of each paper are difficult to understand, and the volume of the papers is also very large. Therefore, it is difficult to quickly determine whether the paper in question is useful for cancer genomic medicine and which part of the paper contains important information. For this reason, several technologies have been proposed to improve the display of documents that users are trying to refer to.

[0004] For example, the following Patent Document 1 discloses a technique for checking only sentences that are to be checked in long patent information data. Specifically, Patent Document 1 discloses an information processing device in which a processing means has a content-discriminating character search section, a highlight display processing section, a sentence-delimiter determining section, a display determination section, and a non-display setting section, and the display means displays sentences separated by paragraph symbols that the display determination section determines as including the highlighted characters for each case data, and also displays the highlighted characters that have been colored by the highlight display processing section.

[0005] Furthermore, the following Patent Document 2 discloses a technology for automatically searching for highly relevant medical reports from past medical reports by using, as keywords, words or word strings weighted according to the occurrence frequency and importance of cases in order to automatically search for relevant case reports from the text of past medical reports, including the examination purpose section, the findings section, and the diagnosis section. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] JP 2016-207071 A [Patent Document 2] JP 2012-141797 A Summary of the Invention [Problem to be solved by the invention]

[0007] However, the technology disclosed in the above-mentioned Patent Document 1 requires the user to input the content-discriminating characters he / she wants to highlight from the input means while thinking about them one by one, which is troublesome and cumbersome. Also, the highlighted characters are subjective to the user, and do not necessarily accurately indicate that such highlighted characters are important or closely related in a particular field.

[0008] In addition, in the technology disclosed in the above Patent Document 2, in order to use words or word strings weighted by occurrence frequency and importance as search keywords, doctors and others had to select words or word strings that they considered important in the displayed report and add them to the importance keyword list in advance. Therefore, as in the above, it did not necessarily accurately indicate that the word or word string is important or closely related in a specific field.

[0009] Therefore, the present invention aims to provide a document display support technology that can estimate words that are considered to be important or closely related to a specific field in a document that a user wishes to refer to, and display the estimated words in a visually distinct manner. [Means for solving the problem]

[0010] The present invention, which aims to solve the above problems, includes the following invention-specific matters and technical features.

[0011] According to one aspect, the present invention is a document display support system that may include a database in which a selection target word set consisting of words to be selected in a document in a specific field and a non-selection target word set consisting of words that should not be selected, a trained word selection model that has been subjected to machine learning to estimate a label indicating at least one of a selection target word and a non-selection target word for a given word according to a context related to the word, a document acceptance unit that accepts a display target document, a text pre-processing unit that extracts a plurality of words from the accepted display target document, a word classification unit that classifies each of the plurality of words in the accepted display target document into at least one of a selection target word, a non-selection target word, and an uncertain word based on the database, a text post-processing unit that generates output data by adding predetermined character attribute information to a predetermined word in the display target document, and an output unit that outputs the output data. The trained word selection model may classify the uncertain word into a selection target word when a label indicating the selection target word is estimated for the classified uncertain word. The text post-processing unit may add the predetermined character attribute information to the classified selection target word.

[0012] The word classification unit may classify, for each of the plurality of words, a word that is not classified as either the selection target word or the non-selection target word, as the uncertain word based on the database.

[0013] According to another aspect, the present invention is a document display support system that may include a database in which a plurality of documents are registered, a trained word selection model that has been subjected to machine learning to estimate a label indicating at least one of a selection target word and a non-selection target word for a given word according to a context related to the word, a document receiving unit that receives an input of a display target document, a text pre-processing unit that extracts a plurality of words from the received display target document, a word classification unit that calculates a predetermined score for each of the extracted plurality of words based on the database and classifies the extracted words into at least one of a selection target word, a non-selection target word, and an uncertain word according to the calculated predetermined score, a text post-processing unit that generates output data by adding predetermined character attribute information to the predetermined word in the display target document, and an output unit that outputs the output data. When a label indicating the selection target word is estimated for the classified uncertain word, the trained word selection model classifies the uncertain word into a selection target word. Also, the text post-processing unit adds the predetermined character attribute information to the classified selection target word.

[0014] The database may include a positive example document corpus defined as a correct answer case. The word classification unit may classify a word as the selection target word when a first score calculated for the word using the positive example document corpus exceeds a first threshold.

[0015] The database may further include a negative example document corpus defined as incorrect answers. The word classifier may calculate the first score based on a total number of documents in which the word appears in all the positive example documents of the positive example document corpus and all the negative example documents of the negative example document corpus relative to a number of documents in which the word appears in all the positive example documents of the positive example document corpus.

[0016] The word classification unit may classify the word as the non-selection target word based on a second score according to a frequency of a document in which the word appears, for all positive example documents in the positive example document corpus and all negative example documents in the negative example document corpus. For example, the word classification unit may classify the word as the non-selection target word when a second score according to a predetermined inverse document frequency does not exceed a second threshold value, based on all positive example documents in the positive example document corpus and all negative example documents in the negative example document corpus. Alternatively, the word classification unit may classify the word as the non-selection target word when a second score according to a predetermined document frequency exceeds a second threshold value.

[0017] In addition, the document display support system may further include a trained document classification model that has been subjected to machine learning to calculate the accuracy of documents based on the frequency of occurrence of words in the specific field, using as training data a first group of documents in a specific field document corpus related to the specific field and a second group of documents in a general field document corpus in a field broader than the specific field.

[0018] The document display support system may further include a document extraction unit that randomly extracts the second document group from the general field document corpus so that the number of documents in the second document group is substantially the same as that of the first document group.

[0019] The trained document classification model can accept a document to be classified and calculate the accuracy of the accepted document to be classified, thereby predicting whether the document is a correct document or an incorrect document based on a predetermined threshold value.

[0020] The above-mentioned text post-processing unit may generate the output data configured so that the predetermined words to which the predetermined character attribute information has been assigned are displayed in a visually distinct manner.

[0021] The present invention according to another aspect may be a document display support method for displaying a predetermined word in a document to be displayed so as to be visually distinguished. The method may include: registering a selection target word set consisting of words to be selected in a document in a specific field and a non-selection target word set consisting of words not to be selected in a database; constructing a trained word selection model in which machine learning is performed for estimating a label indicating at least one of a selection target word and a non-selection target word for a given word according to a context related to the word; extracting a plurality of words from the received document to be displayed; classifying each of the plurality of words in the received document to be displayed into at least one of a selection target word, a non-selection target word, and an uncertain word based on the database; generating output data by assigning predetermined character attribute information to the predetermined word in the document to be displayed; and outputting the output data. The trained word selection model may classify the uncertain word into a selection target word when a label indicating the selection target word is estimated for the classified uncertain word. Furthermore, generating the output data may include assigning the predetermined character attribute information to the classified selection target word.

[0022] The present invention according to another aspect may be a document display support method for displaying a predetermined word in a document to be displayed so as to be visually distinguished. The method may include: constructing a trained word selection model that has been subjected to machine learning for estimating a label indicating at least one of a selection target word and a non-selection target word according to a context related to a given word; extracting a plurality of words from the received document to be displayed; calculating a predetermined score for each of the extracted plurality of words based on a database in which a plurality of documents are registered, and classifying the extracted words into at least one of a selection target word, a non-selection target word, and an uncertain word according to the calculated predetermined score; generating output data by assigning predetermined character attribute information to the predetermined word in the document to be displayed; and outputting the output data. When a label indicating the selection target word is estimated for the classified uncertain word, the trained word selection model may classify the uncertain word into a selection target word. Furthermore, generating the output data may include assigning the predetermined character attribute information to the classified selection target word.

[0023] In addition, the method may further include constructing the trained document classification model that has been subjected to machine learning to calculate the accuracy of documents based on the frequency of occurrence of words in the specific field, using as training data a first group of documents in a specific field document corpus related to the specific field and a second group of documents in a general field document corpus in a field broader than the specific field.

[0024] Furthermore, the present invention can be understood as a product including a computer program for implementing the above-described method by a computing device and a recording medium on which the computer program is non-transitoryly recorded.

[0025] In this disclosure, the term "means" does not simply mean a physical means, but also includes cases where the functions of the means are realized by software. Also, the functions of one means may be realized by two or more physical means, and the functions of two or more means may be realized by one physical means.

[0026] In addition, in this disclosure, a "system" includes a logical collection of multiple devices (or functional modules that realize a specific function), regardless of whether each device or functional module is physically configured as a single entity or as separate entities. Effect of the Invention

[0027] According to the present invention, for documents in a specific field that a user wishes to refer to, a word selection model is used to estimate words that are considered to be important or closely related in the specific field as selection target words, and the estimated selection target words can be displayed in a visually distinct manner.

[0028] Other technical features, objects, and advantages of the present invention will become apparent from the following embodiments described with reference to the accompanying drawings. [Brief description of the drawings]

[0029] [Figure 1] 1 is a block diagram showing an example of a schematic configuration of a document display support system according to an embodiment of the present invention. [Diagram 2] 1 is a block diagram for explaining a functional configuration of a document display support system according to an embodiment of the present invention. [Figure 3A] FIG. 2 is a diagram for explaining machine learning of a word selection model in the document display assistance system according to an embodiment of the present invention. [Figure 3B] FIG. 13 is a diagram for explaining estimation using a trained word selection model in the document display assistance system according to an embodiment of the present invention. [Figure 4]1 is a diagram showing an example of a document displayed on a browser screen in a document display assistance system according to an embodiment of the present invention; [Diagram 5] 1 is a block diagram for explaining a functional configuration of a document display support system according to an embodiment of the present invention. [Figure 6] 1 is a diagram for explaining an example of a configuration of a document database in a document display assistance system according to an embodiment of the present invention; [Figure 7] 1 is a block diagram showing an example of a schematic configuration of a document display assistance system according to an embodiment of the present invention. [Figure 8] 4 is a flowchart for explaining a process of constructing a document database in the document display assistance system according to an embodiment of the present invention. [Figure 9] 5 is a flowchart for explaining a learning process of a word selection model in the document display assistance system according to an embodiment of the present invention. [Figure 10] 5 is a flowchart for explaining a learning process of a word selection model in the document display assistance system according to an embodiment of the present invention. [Figure 11] FIG. 1 is a diagram illustrating an example of a hardware configuration of a document display assistance system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0030] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. However, the embodiment described below is merely an example, and there is no intention to exclude various modifications and application of techniques not specified below. The present invention can be implemented by various modifications (for example, combining the respective embodiments) within the scope of the gist of the present invention. In addition, in the following description of the drawings, the same or similar parts are represented by the same or similar reference numerals. The drawings are schematic and do not necessarily correspond to actual dimensions, ratios, etc. The drawings may also include parts with different dimensional relationships and ratios.

[0031] In the present disclosure, a document display support system and a document display support method are described for selecting words characteristically indicating the contents of a document to be displayed by a selection process using highly relevant documents selected by a filtering process from a large number of documents in a specific field, and displaying the document so that the selected words are visually distinguishable. As described below, in the filtering process, a document classification model that has been subjected to machine learning using a known machine learning algorithm such as a neural network is used. In addition, in the word selection process, a word selection model that has been subjected to machine learning is used. In this disclosure, the term "word" is not limited to a term or terminology, but broadly refers to a semantic group that has a grammatical function, and is not limited to a single word, but includes a combination of multiple words or a phrase.

[0032] [First embodiment] This embodiment is directed to a document display support system and method that uses a word database in which words related to a specific field are registered to classify each word in the text body of a document into one of a selection target word, a non-selection target word, and an uncertain word, and further uses a learned word selection model that has been subjected to machine learning using a predetermined machine learning algorithm to estimate whether the classified uncertain word is a selection target word or a non-selection target word, and displays a document in which the selection target words classified or estimated in this way are visually distinguished.

[0033] 1 is a block diagram showing an example of a schematic configuration of a document display support system according to an embodiment of the present invention. As shown in the figure, the document display support system 100 includes, for example, a cloud computer 110 and a client computer 120. Here, the document display support system 100 is configured as a cloud-type system, but is not limited to this and may be configured as a client / server system or a standalone system.

[0034] The cloud computer 110 is, for example, a computing device having a function of analyzing a document specified by the client computer 120. The hardware configuration of the cloud computer 110 is shown in, for example, FIG. 11, but since such a configuration is known, details thereof will be omitted. The cloud computer 110 sends the analysis result back to the client computer 120. The cloud computer 110 may be configured to include a word database 210 and a trained word selection model 236 that has been previously subjected to machine learning, as described with reference to FIG. 2. The cloud computer 110 may be configured by one or more computing devices.

[0035] The client computer 120 is a computing device operated by a user of the document display assistance system 100. The client computer 120 may be implemented with, for example, a web browser and access the cloud computer 110 via the web browser. Alternatively, instead of a web browser, a dedicated browser or application adapted to the document display assistance system 100 may be implemented. As an example, the client computer 120 transmits to the cloud computer 110 a document (e.g., the document data itself or the URL of the document data) for which the user wants to determine whether or not it is relevant to a specific field. In response to this, the cloud computer 110 transmits the analysis result of the document back to the client computer 120, and the client computer 120 displays the analysis result on a user interface.

[0036] The document display support system 100 will be described in more detail below. FIG. 2 is a block diagram for explaining the functional configuration of the document display support system according to an embodiment of the present invention. In the figure, the document display support system 100 includes, for example, a word database 210, a document reception unit 220, a document analysis unit 230 having a trained word selection model 236, a text post-processing unit 240, and a document display unit 250. In the cloud-type document display support system 100 as described above, for example, the word database 210, the document reception unit 220, the document analysis unit 230, and the text post-processing unit 240 are realized as functions of the cloud computer 110, and the document display unit 250 is realized as a function of the client computer 120, but is not limited to this. For example, all or a part of the text post-processing unit 240 may be realized as a function of the client computer 120. The document reception unit 220, the document analysis unit 230, and the text post-processing unit 240 form a document processing unit 260.

[0037] The word database 210 is a database system in which a word set to be selected (hereinafter referred to as a "selection target word set") and a word set to be not selected (hereinafter referred to as a "non-selection target word set") are registered, which are classified according to word class classification for documents in a specific field. The word database 210 is one form of a dictionary database. Word class classification refers to, for example, using a corpus of documents defined in advance as correct answer cases (hereinafter referred to as a "positive example document") and a corpus of documents defined in advance as incorrect answer cases (hereinafter referred to as a "negative example document"), scoring each word (for example, on a scale of -100 to 100) using a predetermined statistical model, and classifying words according to the assigned score based on a predetermined threshold value. A correct answer case refers to a case in which a certain document is related to a specific field, and an incorrect answer case refers to a case in which a certain document is not related to the specific field. In other words, a selection target word set is a set of words that the document display support system 100 definitely wants to select in a certain display target document, and a non-selection target word set is a set of words that the document display support system 100 definitely does not want to select. As a result of the word class classification, for example, a set of words having a score that is neither low nor high (for example, a score of 0 or around 0) is an uncertain word set.

[0038] The document receiving unit 220 functions as an interface for receiving a document as a display target document, the relevance of which to a specific field is to be visually determined. The document received by the document receiving unit 220 is displayed by the document display unit 250, which will be described later, with the relevance to the specific field being visually expressed. As an example, the document receiving unit 220 receives a document acquired from a literature database on the Internet. Alternatively, the document receiving unit 220 may receive a document acquired from a local computer. As another example, the document receiving unit 220 may receive a document specified by a user via a Web browser. The document receiving unit 220 passes the received document to the document analysis unit 230 and also passes it to the text post-processing unit 240.

[0039] The document analysis unit 230 analyzes the received document and identifies words in the document that are important in a specific field. The identified words are displayed in a visually distinct manner (e.g., highlighted). As described above, in this example, the document analysis unit 230 includes a text preprocessing unit 232, a word classification unit 234, and a word selection model 236.

[0040] The text pre-processing unit 232 extracts a plurality of words from the text body of the received document. A morphological analysis unit that performs a morphological analysis of a sentence, which is obvious to those skilled in the art, is one form of the text pre-processing unit 232. The text pre-processing unit 232 passes each of the extracted words to the word classification unit 234.

[0041] The word classification unit 234 refers to the word database 210 for each word included in the received document, and identifies and classifies the word as a selection target word, a non-selection target word, or an uncertain word. That is, the word classification unit 234 refers to the word database 210 for the word extracted by the text pre-processing unit 232, and if the word exists in the selection target word set, assigns or associates the word with a learning label (e.g., "1") indicating a selection target word, whereas if the word exists in the non-selection target word set, assigns or associates the word with a learning label (e.g., "-1") indicating a non-selection target word. Furthermore, if the extracted word does not exist in either the selection target word set or the non-selection target word set, the word classification unit 234 classifies the extracted word as an uncertain word. In this example, a learning label is not assigned to the uncertain word. The word classification unit 234 passes each word (word string) classified as a selection target word, a non-selection target word, and an uncertain word to the word selection model 236 and to the text post-processing unit 240. Alternatively, the word classification unit 234 may pass the words classified as uncertain words to the word selection model 236 and pass the words classified as selection target words and non-selection target words to the text post-processing unit 240.

[0042] The word selection model 236 of the present embodiment is a trained model that has undergone machine learning to estimate the label of each word (e.g., whether the word is a selection target word or a non-selection target word) for each word extracted from a text sentence of a certain document while considering the context and semantics related to the word. The context refers to the logical relationship or continuity between individual words in a sentence and / or between individual sentences or phrases. For example, when a word is used with the same meaning, it is the same context, and when a word is used with a different meaning, it is a different context. The context depends on the word string (which may be a sentence or phrase) before and after the word. In this disclosure, the word selection model 236 before or during training may also be simply referred to as the "word selection model 236", but may also be referred to as the "trained word selection model 236" to clearly indicate that it has been trained. For example, a sequence labeling model such as a hidden Markov model (HMM) or a long short-term memory (LSTM) may be applied to the word selection model 236, but is not limited thereto. In machine learning, for example, each word extracted by morphological analysis processing is converted into a word embedding vector. A word embedding vector is a representation of a vocabulary by embedding a vector representing the feature amount of each word in a vector space representing the feature of the vocabulary. In the present disclosure, a word embedding vector is used, but this is not limited, and other feature vectors may be used. As described later, uncertain words classified into an uncertain word set are input to the word selection model 236, and labels for the uncertain words are estimated.

[0043] For example, consider a case where a word selection model 236 is machine-learned using a text sentence "... the c-Myc-Skp2-Miz1 complex that activates RhoA and the c-MycpTEFB complex ..." of a document in the field of cancer genome medicine as training data. As shown in FIG. 3A, the text sentence is segmented into a plurality of words, and each segmented word is converted into a word embedding vector. Each word converted into a word embedding vector is input to the word selection model 236, and the word selection model 236 performs machine learning so that an output that is the same as a learning label given in advance is obtained for words corresponding to selection target words and non-selection target words. In the example shown in FIG. 3A, the words "activates" and "RhoA" are uncertain words, and therefore no learning label is given.

[0044] Returning to FIG. 2, after the above-mentioned machine learning, the word selection model 236 estimates a label for the input word while considering the context (e.g., the string of words before and after the word selection). That is, the word selection model 236 assigns a label of either a selection target word or a non-selection target word to the input word, and outputs the pair of the word and the label as an estimation result. As an example, the words input to the word selection model 236 include, in addition to the uncertain words, words other than the uncertain words (e.g., the selection target words and / or the non-selection targets) as context information. This makes it possible to perform estimation by utilizing the generalization ability of the trained model. As another example, the word to be input may be, for example, a word classified as an uncertain word by the word classification unit 234.

[0045] That is, for a document including the above-mentioned text sentence "... the c-Myc-Skp2-Miz1 complex that activates RhoA and the c-MycpTEFB complex...", the word classification unit 234 classifies the words "activates" and "RhoA" as uncertain words. In this case, as shown in FIG. 3B, the trained word selection model 236 estimates labels for the words "activates" and "RhoA" classified as uncertain words according to the words and their surrounding contexts. FIG. 3B shows that the label "-1" is estimated for the word "activates" as a non-selection target word from its context, and the label "+1" is estimated for the word "RhoA" as a selection target word from its context.

[0046] The text post-processing unit 240 performs modification processing on the text sentence of the received document based on the estimation result output by the word selection model 236 so that the selection target words are displayed in a visually distinguished manner in the received document. That is, the text post-processing unit 240 generates output data for the document delivered from the document receiving unit 220, configured so that the selection target words are displayed in a visually distinguished manner in the text sentence of the document based on the selection target words and non-selection target words delivered from the word classification unit 234 and the selection target words estimated and delivered by the word selection model 236. The visually distinguished manner is typically a highlight display, and includes, but is not limited to, modes such as underlining, coloring, italics, bold, enlargement, frame, and / or highlighting. The output data holds, for example, character attribute information indicating such a manner in association with the selection target words in the text sentence of the document. The text post-processing unit 240 delivers the generated output data to the document display unit 250.

[0047] The document display unit 250 displays the modified document based on the output data delivered from the text post-processing unit 240. As described above, the document display unit 250 is realized, for example, as a part of the functions of the client computer 120. For example, the document display unit 250 interprets the output data delivered from the text post-processing unit 240, and displays the selected words in the text sentence of the document on the user interface in a visually distinguished manner as shown in FIG. 4 according to the character attribute information therein. That is, the figure shows an example of a browser screen 400, and more specifically, shows a state in which a word identified as a selected word is highlighted in a display target document displayed on the browser screen (represented as a boxed character in the figure). In this example, the score of the document, related keywords, etc. are additionally displayed on the right side of the browser screen 400.

[0048] As described above, according to this embodiment, the document display support system 100 displays words related to a specific field in a visually distinct manner for a document for which it is desired to determine whether or not the document is related to a specific field. This allows the user to easily and quickly follow the words displayed in a visually distinct manner, thereby enabling the user to immediately determine whether or not the document is related to a specific field.

[0049] In particular, according to this embodiment, the document display assistance system 100 inputs words classified as uncertain words by the word classification unit 234 to the word selection model 236, which has been subjected to machine learning to calculate the context similarity for a specified word in a document in a specific field using registered selection target word sets and non-selection target word sets, thereby making it possible to more accurately estimate whether a word that was previously unable to be determined as being in a selection target word set related to a specific field is part of the selection target word set.

[0050] Furthermore, according to this embodiment, it is not necessary to manually set or register selection target words related to a specific field in advance, and the user is freed from this hassle.

[0051] [Second embodiment] This embodiment is a modification of the above-described first embodiment, and is directed to a document display support system and a document display support method using a document database configured to include a positive example document corpus and a negative example document corpus.

[0052] FIG. 5 is a block diagram for explaining the functional configuration of a document display support system according to an embodiment of the present invention. In the figure, the same components as those in the above embodiment are given the same reference numerals, and therefore, in the following, their explanation will be omitted as appropriate. That is, as shown in the figure, the document display support system 100 of this embodiment differs from the first embodiment in that a document database 510 is adopted instead of the word database 210 in the first embodiment. In other words, the document database 510 is one aspect of the database 210 shown in FIG. 1. In addition, the word classification unit 234 is configured to operate in accordance with the document database 510.

[0053] The document database 510 includes, for example, a positive example document corpus 512 and a negative example document corpus 514. As described above, the positive example document corpus 512 is a database in which documents defined in advance as correct examples (positive example documents) are registered. Also, the negative example document corpus 514 is a database in which documents defined in advance as incorrect examples are registered.

[0054] As described in other embodiments, such a document database 510 is constructed by inputting a large number of documents to a trained document classification model that has been subjected to machine learning in advance using a document corpus in a specific field that has a small domain but a high frequency of occurrence of words to be selected and a document corpus in a general field that is a broader domain than the domain of the document corpus in the specific field, and classifying each of the input documents into either a positive example document or a negative example document. The document database 510 may accumulate documents that have been collected and classified from a document corpus related to journals, academic papers, etc., for example.

[0055] There are a huge number of documents such as journals and academic papers around the world, and while some can be obtained free of charge from open access databases, many must be purchased from paid databases. Such documents generally include bibliographic information, abstracts, and text, and some databases (e.g., PubMed Central, etc.) make the text available free of charge or at a low cost. Therefore, in this disclosure, it is assumed that the texts of such documents available free of charge or at a low cost are stored in the document database 510 as positive example documents and negative example documents, but this is not intended to be limiting.

[0056] The word classification unit 234 refers to the document database 510 and classifies each word included in the received document as a selection target word, a non-selection target word, or an uncertain word. That is, the word classification unit 234 uses the positive example document corpus 512 and the negative example document corpus 514 to score each word (for example, on a scale of 0 to 100) according to a predetermined statistical model, and classifies the word as a selection target word, a non-selection target word, or an uncertain word according to the assigned score based on a predetermined threshold value. For example, the word classification unit 234 classifies a word that may or may not be a selection target word depending on the context, or a word whose calculated score is a neutral score (for example, near 0, etc.) as an uncertain word, and reserves the classification of whether it is a selection target word or not. As described above, the learned word selection model 236 estimates whether an uncertain word is a selection target word or a non-selection target word.

[0057] In this example, classification of words to be selected according to the scoring by the word classification unit 234 is performed as follows. That is, the word classification unit 234 calculates the score (degree of accuracy) of the extracted word using a scoring function shown in the following formula, for example.

number

number

[0058] In other words, the above formula 1 indicates the ratio of the number of documents N in which a certain word appears in all documents D (i.e., the positive example document corpus 512 and the negative example document corpus 514) to the number of documents N+ in which the certain word appears in the positive example documents D+ of the positive example document corpus 512 (for example, if the certain word appears (exists) even once in a certain document, +1 is counted).

[0059] For example, consider the score of the word t1 "EGFR" (Epidermal Growth Factor Receptor) and the word t2 "the" contained in a document in a particular field. For simplicity, α=0.

[0060] 6, the document database 510 stores an entire document D consisting of an entire positive example document D+ and an entire negative example document D−. A document d in the entire positive example document D+ and the entire negative example document D− includes “EGFR” and “the”.

[0061] From Equation 1, the number of positive documents containing word t, N+, is given by N+=|d∈D+:t∈d| The number of documents N that contain word t in all documents D is N = |d∈D:t∈d| Therefore, the score of the word t1 "EGFR" is

number

[0062] Similarly, the score for word t2 "the" is

number

[0063] The word classification unit 234 calculates a score for each of the extracted words as described above, and identifies and classifies (labels) words whose calculated scores exceed the first threshold value (in the above example, word t1 “EGFR”) as selection target words.

[0064] For example, for a document in the field of cancer genomic medicine, the word classifier 234 may identify and classify the following words as selected target words: L858R / E545K / H1047R / T790M / EGFR-TKIs / EGFR-TKI / G12C / G13D / MuTect / GISTIC / E542K / ASXL1 / FLT3-ITD / V600E / NRAS / G12V / …

[0065] On the other hand, the classification of non-selection candidate words by the word classification unit 234 is performed as follows: That is, the word classification unit 234 calculates the score (incorrectness) of the extracted word by using, for example, the inverse document frequency function (idf) shown in the following formula:

number

[0066] Similarly, the scores based on the inverse document frequency function are considered for all the positive example documents D+ and all the negative example documents D− shown in FIG.

[0067] That is, from Equation 5, the score of the word t1 “EGFR” according to the inverse document frequency function is

number

number

[0068] In this example, the inverse document frequency function idf is used, but the document frequency function df may be used instead. In this case, words whose calculated scores exceed a predetermined threshold value are identified and classified as non-selection candidates.

[0069] In this way, the word classification unit 234 scores each word according to a predetermined statistical model, and classifies the word into one of a selection target word, a non-selection target word, and an uncertain word according to the assigned score based on a predetermined threshold value. Then, as described above, the word classification unit 234 passes the words classified into the selection target word and the non-selection target word to the text post-processing unit 240, and passes each word classified into the selection target word, the non-selection target word, and the uncertain word to the word selection model 236. As a result, as described above, the word selection model 236 estimates the label of the input uncertain word (i.e., whether the word is a selection target word or not) while taking into account the context.

[0070] As described above, according to this embodiment, it is possible to obtain the same advantages as the first embodiment, and by using the body text of documents that are available free of charge or at low cost as positive example documents and negative example documents, it is possible to construct a document display support system 100 efficiently and at low cost.

[0071] [Third embodiment] This embodiment is a modification of the second embodiment, and is directed to a method for constructing a document database including a positive example document corpus and a negative example document corpus.

[0072] FIG. 7 is a block diagram showing an example of a schematic configuration of a document display support system according to an embodiment of the present invention. In the figure, the same components as those in the above embodiment are given the same reference numerals, and therefore, in the following, their description will be omitted as appropriate. That is, as shown in the figure, the document display support system 100 of this embodiment is different from the configuration shown in the above second embodiment in that a document selection unit 710 is added. In addition, in the figure, the document reception unit 220, the document analysis unit 230, and the text post-processing unit 240 shown in FIG. 2 and the like are simply shown as a document processing unit 260. As shown in the figure, the document selection unit 710 can be configured to include a document extraction unit 712 and a document classification model 714.

[0073] The specific field document corpus 720 is a corpus of documents in which words to be selected have a high frequency of appearance in documents in a specific field or a specific domain. As an example, when the positive example document corpus 512 is constructed so that words related to the cancer genome field are selected, papers referenced from "OncoKB" or "COSMIC," which are knowledge databases related to the cancer genome field, may be selected. In the present disclosure, documents registered in the specific field document corpus 720 are, for example, abstracts of documents such as journals and academic papers, and may not include the main text. The specific field document corpus 720 has a smaller number of registered documents (i.e., abstracts) (e.g., about 15,000 documents) than the general field document corpus 730 described later.

[0074] The general field document corpus 730 is a document corpus of a general field, which is a domain broader than that of the specific field document corpus 720. As an example, the general field document corpus 730 targets abstracts of papers provided by the free search engine "PubMed" in the fields of life science and biomedical science, including the field of cancer genomes. Since the general field document corpus 730 has a huge number of documents (e.g., about 40 million documents), a predetermined number of documents can be randomly extracted by the document extraction unit 712.

[0075] The document extraction unit 712 randomly extracts a predetermined number of documents (i.e., abstracts) (e.g., substantially the same number as the number of documents registered in the specific field document corpus 720) with reference to the general field document corpus 730. By making the number of documents from the general field document corpus 730 substantially the same as the number of documents registered in the specific field document corpus 720, bias in the trained model can be prevented. The document extraction unit 712 inputs the extracted documents to the document classification model 714 as training data. Note that "substantially the same number" does not mean a completely identical number between the two, but a difference in number that is obvious to a person skilled in the art and does not affect the training results.

[0076] The document classification model 714 is a trained model that has undergone machine learning to calculate the accuracy of a document according to the occurrence frequency of words in a specific field, using a group of documents registered in the specific field document corpus 720 and a group of documents extracted from the general field document corpus 730 via the document extraction unit 712. In the present disclosure, the document classification model 714 uses "Bag of Words" to convert words included in the text sentences of documents into feature vectors, and performs machine learning using a known machine learning algorithm. Bag of Words is known as a method of expressing data described in natural language, that is, text sentences, as vectors in natural language processing. In other words, the document classification model 714 calculates the occurrence frequency of each word that appears in each document registered in the specific field document corpus 720, while calculating the occurrence frequency of each word that appears in each general document extracted by the document extraction unit 712, and performs feature vectorization of words according to the calculated occurrence frequency, thereby forming a trained model. Note that, instead of Bag of Words, other feature vectors such as the above-mentioned word embedding vectors may be used.

[0077] After the above-described machine learning, the document classification model 714 calculates the accuracy of the document based on the input document group, thereby estimating whether the document is a correct document or an incorrect document. That is, the document classification model 714 compares the calculated accuracy with a predetermined threshold, and if the calculated accuracy exceeds the predetermined threshold, classifies the document as a correct document and registers it in the positive example document corpus 512. On the other hand, if the calculated accuracy does not exceed the predetermined threshold, the document classification model 714 classifies the document as an incorrect document and registers it in the negative example document corpus 514. In the present disclosure, the document input to the trained document classification model 714 is, for example, the text body of the document, and may not include bibliographic information or an abstract. More specifically, the document may be, for example, the body of a document referenced by PubMed Central (PMC). In addition, the number of documents (text bodies) input to the trained document classification model 714 is very large compared to the number of documents (abstracts) input for machine learning, and may be, for example, one million or more.

[0078] In this way, the document selection unit 710 uses the trained document classification model 714 to classify a very large group of documents into either positive example documents or negative example documents, thereby constructing a document database 510 including a positive example document corpus 512 and a negative example document corpus 514.

[0079] As described above, this embodiment provides the same advantages as the above-mentioned embodiment. In particular, this embodiment uses a limited number of documents and parts of them (such as abstracts) as learning data, so that the document classification model 714 can be efficiently machine-learned. In particular, the documents as learning data are documents extracted from a document corpus in which words to be selected related to a specific domain have a high frequency of appearance, and a document corpus in a general field, which is a broader domain than the specific domain, so that the document classification model 714 can efficiently learn whether a document is a positive example document or a negative example document.

[0080] Furthermore, according to this embodiment, the machine learning-based document classification model 714 can calculate the accuracy of an input document according to the frequency of occurrence of words in a specific field, and can therefore determine whether the input document is a correct document or an incorrect document based on a predetermined threshold value.

[0081] Furthermore, according to this embodiment, it is possible to correctly infer, as a selection target word, a word that is not registered in a conventional dictionary database but should be determined as a selection target word in a document in a specific field. For example, in the field of cancer genome medicine, a word representing "mutation" (e.g., written as "L747S") is described in an extremely large number of patterns in documents in the field. Therefore, such a word is practically no longer compatible with a conventional dictionary database, but according to this embodiment, it is possible to infer it as a selection target word.

[0082] Next, a series of processing steps in the document display assistance system 100 shown in each of the above embodiments will be described with reference to FIGS.

[0083] 8 is a flowchart for explaining a process for constructing a document database in the document display assistance system 100 according to an embodiment of the present invention. This process is realized, for example, by the cloud computer 110 executing a predetermined program under the control of a processor.

[0084] As shown in the figure, in the document display assistance system 100, the document selection unit 710 acquires a predetermined number of documents from the specific field document corpus 720 (S801), and acquires substantially the same number of documents from the general field document corpus 730 (S802). As described above, the documents acquired here may be a portion of a document, for example, an abstract. In this way, by acquiring substantially the same number of documents from the specific field document corpus 720 and the general field document corpus 730, it is possible to prevent bias in the trained model.

[0085] Next, the pre-learning document classification model 714 performs machine learning to calculate the accuracy of the document according to the frequency of occurrence of words in a specific field based on each of the acquired documents (S803). In the machine learning, for example, the frequency of occurrence of words in a text sentence is calculated and vectorized using Bag of Words.

[0086] After constructing the trained model, the document classification model 714 accepts an input of a document to be classified and calculates the accuracy of the document (S804). The input document is a part of the document, for example, the body text. The document classification model 714 compares the calculated accuracy of the document with a predetermined threshold value and determines whether the accuracy of the document exceeds the predetermined threshold value (S805).

[0087] If the document classification model 714 determines that the accuracy of the document exceeds the predetermined threshold (Yes in S805), it registers the document as a correct document in the positive example document corpus 512 (S806). On the other hand, if the document classification model 714 determines that the accuracy of the document does not exceed the predetermined threshold (No in S805), it registers the document as an incorrect document in the negative example document corpus 514 (S807).

[0088] The document sorting unit 710 then determines whether all documents to be sorted have been sorted (S808), and if it determines that there are unsorted documents (No in S808), it returns to the process of S804 and repeats the process until all documents to be sorted have been sorted. On the other hand, if the document sorting unit 710 determines that all documents to be sorted have been sorted (Yes in S808), it ends the document database construction process.

[0089] 9 is a flowchart for explaining a learning process of a word selection model in the document display assistance system 100 according to an embodiment of the present invention. This process is realized, for example, by the cloud computer 110 executing a predetermined program under the control of a processor.

[0090] As shown in the figure, the word selection model 236 accepts the positive example documents of the positive example document corpus 512 and the negative example documents of the negative example document corpus 514 of the document database 510 constructed according to the above-mentioned process as learning data (S901). Next, the word selection model 236 performs machine learning for each word in the positive example document and the negative example document while taking into account the context of the word (S902) to enable estimation of the label of the word. The word selection model 236 thus subjected to machine learning becomes capable of estimating and outputting a label for a given word input.

[0091] 10 is a flowchart for explaining the document display support process in the document display support system 100 according to an embodiment of the present invention. This process is realized, for example, by the cloud computer 110 executing a predetermined program under the control of a processor.

[0092] That is, as shown in the figure, the document receiving unit 220 receives a document to be displayed, the document being visually judged as to whether it is related to a specific field (S1001). Next, the text preprocessing unit 232 extracts a plurality of words from the text body of the received document, for example, by using a morphological analysis process or the like (S1002).

[0093] Next, the word classification unit 234 calculates a score for the extracted word based on a predetermined scoring function with reference to the document database 510 (S1003). For example, the calculated score is a degree of correctness (for example, a value between 0 and 100) and a degree of incorrectness (for example, a value between -100 and 0). The word classification unit 234 compares the calculated score with a predetermined threshold value (S1004) to classify the word. That is, if the calculated score (however, a positive value) for the word exceeds the first threshold value, the word classification unit 234 classifies the word as a selection target word (S1005). On the other hand, if the calculated score (however, a negative value) for the word does not exceed the second threshold value, the word classification unit 234 classifies the word as a non-selection target word (S1006). In addition, the word classification unit 234 classifies the word as an uncertain word in cases other than the above, that is, if the calculated score does not exceed the first threshold value and exceeds the second threshold value (S1007).

[0094] Next, the word selection model 236 estimates whether the uncertain word is a selection target word or not based on the input words (e.g., a selection target word, a non-selection target word, and an uncertain word) and in accordance with the surrounding context (S1008). As a result of such estimation, the word selection model 236 assigns a label of either a selection target word or a non-selection target word to the input word, for example.

[0095] After the word selection model 236 is estimated, the text post-processing unit 240 performs modification processing on the text sentences of the received document based on the estimation result output by the word selection model 236 so that the selection target words are displayed in a visually distinct manner in the document (S1009). As an example, the text post-processing unit 240 generates output data in which character attribute information is added to the document so that the selection target words are displayed in a highlighted manner. Then, the document display unit 250 displays the modified document on the user's interface based on the output data generated by the text post-processing unit 240.

[0096] As described above, according to the embodiment described above, the document display support system 100 uses the word selection model 236 to estimate, for a document in a specific field that the user wishes to refer to, words that are considered to be important or closely related in the specific field as selection target words, and can display the estimated selection target words in a visually distinct manner.

[0097] The above-described embodiments are merely examples for explaining the present invention, and are not intended to limit the present invention to these embodiments. The present invention can be embodied in various forms without departing from the spirit of the present invention.

[0098] For example, in the present disclosure, the cancer genome field has been described as an example of a specific field, but the specific field is not limited to this. Furthermore, documents may include not only papers, but also reports, tables, electronic medical records, and the like.

[0099] Furthermore, the above-mentioned document display support system 1 is configured to visually distinguish and display the selection target words in the document, but is not limited to this, and may be configured to further utilize the selection target words, for example. That is, the words classified and highlighted as selection target words in the document to be displayed by the document display support system 1 indicate that they are semantically important words in the document. Therefore, the selection target words may be used as meta information (tags) or keywords of the document.

[0100] As an example, for a certain document, the document display support system 1 associates words classified as selection target words by the word selection model 236 with the document as tags and / or keywords and registers them in a database. When a user refers to the document, the document display support system 1 provides the user with a preview of the words associated in this way, thereby enabling the user to quickly grasp the contents of the document.

[0101] As another example, the document display assistance system 1 may have a function of suggesting a search query based on selection target words registered in association with a document, or a function of weighting search results.

[0102] As yet another example, the document display support system 1 may have a function of identifying parts of a document, such as sentences, paragraphs, or chapters, in which a large number of selection target words appear, and generating a summary of the document using natural language processing centered on the identified parts.

[0103] As yet another example, the document display assistance system 1 may be configured to read out the selection target words aloud, for example, in addition to or instead of a configuration in which the selection target words in the document are visually distinguished and displayed.

[0104] In addition, in the methods disclosed herein, steps, operations, or functions may be performed in parallel or in a different order, provided that the results are not inconsistent. The steps, operations, and functions described are provided merely as examples, and some of the steps, operations, and functions may be omitted or combined into one, or other steps, operations, or functions may be added, without departing from the spirit of the invention.

[0105] In addition, although various embodiments are disclosed in this specification, specific features (technical matters) in one embodiment can be added to or replaced with specific features in other embodiments, with appropriate modifications, and such forms are also included in the spirit of the present invention. [Explanation of symbols]

[0106] 100...Document display support system 110…Cloud Computer 120...Client computer 210…Word database 220…Document Reception Department 230…Document Analysis Department 232...Text preprocessing section 234…Word Classification Section 236… Word Selection Model 240…Text post-processing section 250…Document display section 260…Document processing unit 510…Document database 512…Positive document corpus 524…Negative example document corpus 710…Document Selection Department 712…Document Extraction Unit 714…Document classification model 720…Specialized Document Corpus 730…General field document corpus

Claims

1. a database in which a selection target word set consisting of words that should be selected in documents in a specific field and a non-selection target word set consisting of words that should not be selected; a trained word selection model that has been subjected to machine learning to estimate a label indicating at least one of a selection target word and a non-selection target word for a given word according to a context related to the word; a document receiving unit for receiving a document to be displayed; a text preprocessing unit that extracts each of a plurality of words from the received document to be displayed; a word classification unit that classifies each of a plurality of words in the received document to be displayed into at least one of a selection target word, a non-selection target word, and an uncertain word based on the database; a text post-processing unit that generates output data by adding predetermined character attribute information to predetermined words in the document to be displayed; an output unit that outputs the output data, the learned word selection model classifies the uncertain word into a selection target word when a label indicating the selection target word is estimated for the classified uncertain word; the text post-processing unit assigns the predetermined character attribute information to the classified selection candidate words; Document display support system.

2. the word classification unit classifies, for each of the plurality of words, a word that is not classified as either the selection target word or the non-selection target word based on the database, into the uncertain word; 2. The document display support system according to claim 1.

3. A database in which a plurality of documents are registered; a trained word selection model that has been subjected to machine learning to estimate a label indicating at least one of a selection target word and a non-selection target word for a given word according to a context related to the word; a document receiving unit for receiving an input of a document to be displayed; a text preprocessing unit that extracts each of a plurality of words from the received display target document; a word classification unit that calculates a predetermined score for each of the extracted words based on the database, and classifies the extracted words into at least one of selection target words, non-selection target words, and uncertain words in accordance with the calculated predetermined score; a text post-processing unit that generates output data by adding predetermined character attribute information to predetermined words in the document to be displayed; an output unit that outputs the output data, the learned word selection model classifies the uncertain word into a selection target word when a label indicating the selection target word is estimated for the classified uncertain word; the text post-processing unit assigns the predetermined character attribute information to the classified selection candidate words; Document display support system.

4. The database includes a corpus of positive example documents defined as correct examples; the word classification unit classifies the word into the selection candidate word when a first score calculated for the word using the positive example document corpus exceeds a first threshold value; The document display support system according to claim 3.

5. The database further includes a corpus of negative example documents defined as false positives; the word classification unit calculates the first score based on a total number of documents in which the word appears in all the positive example documents of the positive example document corpus and all the negative example documents of the negative example document corpus relative to a number of documents in which the word appears in all the positive example documents of the positive example document corpus; The document display support system according to claim 4.

6. the word classification unit classifies the word into the non-selection candidate word based on a second score according to a frequency of a document in which the word appears for all the positive example documents of the positive example document corpus and all the negative example documents of the negative example document corpus; The document display support system according to claim 5.

7. the word classification unit classifies the word as the non-selection target word when a second score according to a predetermined inverse document frequency does not exceed a second threshold based on all the positive example documents of the positive example document corpus and all the negative example documents of the negative example document corpus, or classifies the word as the non-selection target word when the second score according to a predetermined document frequency exceeds a second threshold. The document display support system according to claim 6.

8. The document classification system further includes a trained document classification model that has been subjected to machine learning to calculate the accuracy of a document according to the frequency of occurrence of a word in the specific field, using a first group of documents in a specific field document corpus related to the specific field and a second group of documents in a general field document corpus in a field broader than the specific field as training data. The document display support system according to any one of claims 1 to 7.

9. a document extraction unit that randomly extracts the second document group from the general field document corpus so that the number of documents in the second document group is substantially the same as the number of documents in the first document group; The document display support system according to claim 8.

10. The trained document classification model receives a document to be classified, calculates the accuracy of the received document to be classified, and estimates whether the document is a correct document or an incorrect document based on a predetermined threshold value.

10. A document display support system according to claim 8.

11. the text post-processing unit generates the output data configured so that the predetermined words to which the predetermined character attribute information has been assigned are displayed in a visually distinct manner. The document display support system according to any one of claims 1 to 10.

12. 1. A document display support method for displaying a document to be displayed so that a predetermined word in the document to be displayed is visually distinguishable, comprising: A selection target word set consisting of words that should be selected in documents in a specific field and a non-selection target word set consisting of words that should not be selected are registered in a database; constructing a trained word selection model that has been subjected to machine learning to estimate a label indicating at least one of a selection target word and a non-selection target word for a given word according to a context related to the word; Extracting a plurality of words from the received document to be displayed; classifying each of a plurality of words in the received document to be displayed into at least one of a selection target word, a non-selection target word, and an uncertain word based on the database; generating output data by adding predetermined character attribute information to predetermined words in the document to be displayed; outputting the output data; the learned word selection model classifies the uncertain word into a selection target word when a label indicating the selection target word is estimated for the classified uncertain word; generating output data includes adding the predetermined character attribute information to the classified selection candidate words; Document display support method.

13. 1. A document display support method for displaying a document to be displayed so that a predetermined word in the document to be displayed is visually distinguishable, comprising: constructing a trained word selection model that has been subjected to machine learning to estimate a label indicating at least one of a selection target word and a non-selection target word for a given word according to a context related to the word; extracting a plurality of words from the received document to be displayed; calculating a predetermined score for each of the extracted words based on a database in which a plurality of documents are registered, and classifying the extracted words into at least one of a selection target word, a non-selection target word, and an uncertain word according to the calculated predetermined score; generating output data by adding predetermined character attribute information to predetermined words in the document to be displayed; outputting the output data; the learned word selection model classifies the uncertain word into a selection target word when a label indicating the selection target word is estimated for the classified uncertain word; generating the output data includes adding the predetermined character attribute information to the classified selection candidate words; Document display support method.

14. and constructing the trained document classification model, which is subjected to machine learning for calculating the accuracy of documents according to the frequency of occurrence of words in the specific field, using a first group of documents in a specific field document corpus related to the specific field and a second group of documents in a general field document corpus in a field broader than the specific field as training data. The document display support method according to claim 13.

15. A computer program product for causing a computing device to execute the document display support method according to any one of claims 12 to 14 under the control of a processor of the computing device.

Images