JP6615392B1 - Information processing apparatus, information processing method, and program - Google Patents

Abstract

Whether or not each of one or more sentences is a future expression sentence including a thing related to the future, and is a future-oriented sentence that is a target means sentence that is a sentence having at least one of a purpose expression and a means expression. Provided are an information processing apparatus, an information processing method, and a program capable of performing future-oriented sentence determination processing for determining whether or not. Whether an information processing apparatus is a future-oriented sentence that is a future expression sentence including a thing related to the future and is a target means sentence that is a sentence having one or more of a purpose expression and a means expression. A detection unit that determines whether or not, an acquisition unit that acquires future-oriented sentence information that is information related to a determination result in the detection unit, and an output unit that outputs the future-oriented sentence information acquired by the acquisition unit. [Selection] Figure 1

Description

  The present invention relates to an information processing apparatus or the like that detects a future-oriented sentence that is a sentence expressing the future and has at least one of a purpose expression and a means expression.

Conventionally, there has been a document feature extraction apparatus that extracts a feature of a target document that can obtain high accuracy when classifying documents using machine learning (see Patent Document 1).
Such an apparatus includes an input unit that inputs a document, a first boundary dividing unit that analyzes the document and divides it into first units, and a document that is divided into first units by the first boundary dividing unit, A second boundary dividing unit that divides each second unit longer than the first unit, and a first unit adjacent to each other within a range of each second unit divided by the second boundary dividing unit. A document feature comprising: a column extracting unit that extracts a first feature together with a first unit before connection as a column; and a feature output unit that outputs the first feature as a document feature It is an extraction device.

Japanese Patent Laid-Open No. 2006-224847

  However, in the prior art, each of the one or more sentences is a future expression sentence including things related to the future, and is a future-oriented sentence that is a target means sentence that is a sentence having one or more of a purpose expression or a means expression. Future-oriented sentence determination processing to determine whether or not there is difficult.

  For example, securities reports include future or current ongoing measures aimed at business performance expansion and risk management. Regarding the management policy, we changed the “issues to be addressed” from the securities report for the fiscal year ending on or after March 31, 2017 to “management policies, management environment and issues to be addressed”, etc. Is also to be described. Therefore, these sentences are expected to reflect the future performance and corporate structure of the company.

  And the future-oriented sentence, which includes the purpose expression and means expression, and expresses the future, expresses that the company will work on the purpose to be achieved in the future using concrete means. It is assumed that By extracting such a sentence, it is thought that it is possible to efficiently grasp the issues recognized by the company, the efforts against those issues and their progress. Furthermore, for example, by gathering this information from multiple companies, it is possible to predict changes in financial indicators, determine the stability of future performance, profitability, and whether governance is good or bad. Conceivable.

  In spite of the above situation, as described above, in the prior art, for example, there is no technology for detecting a future-oriented sentence from a document such as a securities report.

  The information processing apparatus according to the first aspect of the present invention is a future-oriented sentence that is a future expression sentence including a sentence related to the future, and is a target means sentence that is a sentence having at least one of a purpose expression and a means expression. An information processing unit comprising: a detection unit that determines whether or not, an acquisition unit that acquires future-oriented text information that is information related to a determination result in the detection unit, and an output unit that outputs the future-oriented text information acquired by the acquisition unit Device.

  With this configuration, it is possible to perform future-oriented sentence determination processing for determining whether a sentence is a future-oriented sentence.

  Further, the information processing apparatus according to the second aspect of the present invention further includes a sentence information storage unit that stores one or more pieces of sentence information including two or more sentences, with respect to the first invention, A purpose that is a future expression sentence including things related to the future from one or more sentences included in one or more pieces of text information stored in the information storage unit, and a sentence having one or more of a purpose expression or means expression A future-oriented sentence that is a means sentence is detected, and the acquisition unit is an information processing apparatus that acquires future-oriented sentence information that is information related to the future-oriented sentence detected by the detection unit.

  With this configuration, it is possible to perform future-oriented sentence determination processing for each sentence included in one or more pieces of sentence information, and use the result to output information.

  Further, in the information processing apparatus according to the third aspect of the invention, in contrast to the second aspect of the invention, the detection unit includes a future expression sentence detection unit that determines whether the sentence to be examined is a future expression sentence, The target means sentence detection means for determining whether or not the target sentence is a target means sentence, the future expression sentence detection means determines that it is a future expression sentence, and the purpose means sentence detection means is the target means sentence. The acquisition unit is an information processing apparatus that acquires future-oriented sentence information that is information related to the future-oriented sentence selected by the selection means.

  With this configuration, it is possible to perform future-oriented sentence determination processing for each sentence included in one or more pieces of sentence information, and use the result to output information.

  The information processing apparatus according to the fourth aspect of the invention includes an exclusion condition storage unit that stores, for the second or third invention, an exclusion condition that is a condition for removing a sentence not corresponding to a future-oriented sentence, It further judges whether one or more sentences which one or more each sentence information has satisfy | fills an exclusion condition, and further comprises the exclusion part which excludes the sentence which satisfy | fills an exclusion condition, and a detection part is 1 or more each sentence information Is a future expression sentence including matters related to the future from one or more sentences excluding the sentence excluded by the exclusion unit, and having any one or more of the purpose expression or the means expression. It is an information processing device that detects a future-oriented sentence that is a certain purpose means sentence.

  With this configuration, future-oriented sentence determination processing can be performed at high speed and with high accuracy.

  Further, in the information processing apparatus according to the fifth aspect of the invention, in contrast to the fourth aspect, the exclusion condition is a character number condition relating to the number of characters in the sentence, an item in which the sentence appears, an item condition relating to the item corresponding to the sentence, This is an information processing apparatus that is one or more of the attribute conditions related to the attribute of the character, the character conditions related to a specific character included in the sentence, or a combination of two or more conditions.

  With this configuration, future-oriented sentence determination processing can be performed at high speed and with high accuracy.

  The information processing apparatus according to the sixth aspect of the present invention further comprises a sentence ending condition storage unit that stores a sentence ending condition that is a condition related to sentence ending expression, as compared with any one of the first to fifth inventions, and a detection unit Is an information processing apparatus that determines that a sentence that meets the sentence end condition is a future expression sentence or is not a future expression sentence.

  With this configuration, it is possible to detect the future expression sentence at high speed and with high accuracy.

  In addition, the information processing apparatus according to the seventh aspect of the present invention is a future expression sentence with respect to any one of the first to sixth inventions by using two or more future expression sentences that are future expression sentences. The first learner storage unit that stores the first learner, which is information used when determining whether the information is created by a machine learning algorithm, and the detection unit uses the first learner Thus, the information processing apparatus includes first machine learning means for determining whether or not a sentence to be examined is a future expression sentence by a machine learning algorithm.

  With this configuration, it is possible to detect the future expression sentence with high accuracy.

  Further, in the information processing apparatus according to the eighth aspect of the invention, in contrast to the sixth aspect, the first learning device is a vector obtained using two or more future representation correct example sentences, and has a predetermined sentence end. End-of-sentence clue expression information related to clue expression, frequent term information related to frequent terms that are predetermined frequent terms, adverb information about adverb terms included, pair information about pairs of two clauses at the end of sentences, part of speech about morphemes of morphemes at the end of sentences A learning device obtained by learning with an algorithm of machine learning using two or more vectors having one or more pieces of information, phrase information about the phrases included in the sentence, and sentence attribute information about sentence attributes as elements. The first machine learning means, from the sentence to be examined, includes sentence end clue expression information, frequent term information, adverb information, pair information, part of speech information, punctuation information, sentence attribute information Get one or more pieces of information, construct a vector with one or more pieces of information as elements, apply the vector to the first learner, and get information indicating whether it is a future expression sentence by machine learning algorithm Information processing apparatus.

  With this configuration, it is possible to detect the future expression sentence with higher accuracy.

  Further, the information processing apparatus according to the ninth aspect of the present invention is a target means sentence with respect to any one of the first to seventh inventions by using two or more target means correct example sentences that are target means sentences. The second learner storage unit for storing the second learner, which is information used when determining whether or not is the information created by the machine learning algorithm, and the detection unit uses the second learner An information processing apparatus comprising second machine learning means for determining whether or not one or more sentences included in one or more pieces of sentence information are target means sentences by a machine learning algorithm.

  With this configuration, the target means sentence can be detected with high accuracy.

  Further, in the information processing apparatus according to the tenth aspect of the invention, in contrast to the ninth aspect, the second learner is a vector acquired using two or more objective means positive example sentences, and the word before the reading Part-of-speech information before punctuation, part-of-speech part-of-speech information about part-of-speech of the word before the punctuation point, punctuation information about the punctuation included in the sentence, sentence attribute information about the sentence attribute, and future expression sentence acquired by the first machine learning means It is a learning device obtained by learning with an algorithm of machine learning using two or more vectors having one or more pieces of information of first determination result information as to whether or not there is, second machine learning The means acquires one or more pieces of information from one or more sentences included in one or more pieces of sentence information, including part-of-speech information before punctuation, part-of-speech information before punctuation, punctuation information, sentence attribute information, and first determination result information. A vector with one or more pieces of information as elements Configured to apply the vector to the second learning device, the algorithm of machine learning, an information processing apparatus that acquires information indicating whether an object means statement.

  With this configuration, the target means sentence can be detected with higher accuracy.

  In addition, in the information processing apparatus according to the eleventh aspect of the present invention, with respect to any one of the second to tenth aspects, the acquisition unit performs statistical processing using one or more future-oriented sentences detected by the detection unit. The information processing apparatus acquires future-oriented sentence information that is a statistical processing result.

  With this configuration, it is possible to perform statistical processing using the results of future-oriented sentence determination processing.

  Further, in the information processing apparatus of the twelfth aspect of the invention, in contrast to the eleventh aspect, one or more pieces of sentence information in the sentence information storage unit are associated with sentence attribute values, and the acquisition unit This is an information processing apparatus that acquires future-oriented sentence information that is a statistical processing result using an attribute value.

  With this configuration, it is possible to perform statistical processing using the results of future-oriented sentence determination processing.

  Further, in the information processing apparatus of the thirteenth invention, in contrast to the twelfth invention, the sentence attribute value is a company identifier for identifying a company or a business type identifier for identifying a business type. Alternatively, the information processing apparatus acquires future-oriented sentence information, which is a statistical processing result, using an industry type identifier.

  With this configuration, it is possible to perform appropriate statistical processing using the results of the future-oriented sentence determination processing.

  According to the information processing apparatus of the present invention, the one or more sentences are future expression sentences including things related to the future, and the future is an object means sentence that is a sentence having one or more of the object expression and the means expression. Future-oriented sentence determination processing can be performed to determine whether the sentence is intention-oriented.

Block diagram of information processing apparatus A in Embodiment 1 Figure showing an example of specific clue expression at the end of the sentence The figure which shows the dimensionality etc. of the vector which the 1st machine learning means 3311 comprises The figure which shows the dimensionality etc. of the vector which the 2nd machine learning means 3321 comprises Flow chart for explaining an operation example of the information processing apparatus A Flow chart explaining an example of the detection process Flow chart explaining an example of the exclusion process Flowchart explaining an operation example of the future expression sentence inspection Flow chart for explaining an operation example of the same means inspection Flow chart for explaining an operation example of the same means inspection Flow chart explaining an example of the acquisition process Flow chart explaining an example of the first learning process Flow chart for explaining an example of the second learning process Flow chart explaining an example of the detection process Flow chart explaining an example of the detection process Conceptual diagram explaining an operation example of the information processing apparatus A Figure showing a securities report as an example of the same text information Figure showing examples of sentences removed by the first filtering process Figure showing the results of the experiment Figure showing the results of the experiment The figure which shows evaluation of the same information processing equipment Figure showing example of future-oriented sentence information Overview of the computer system Block diagram of the computer system

  Hereinafter, embodiments of an information processing apparatus and the like will be described with reference to the drawings. In addition, since the component which attached | subjected the same code | symbol in embodiment performs the same operation | movement, description may be abbreviate | omitted again.

(Embodiment 1)
In the present embodiment, an information processing apparatus that determines whether a sentence is a future-oriented sentence, acquires future-oriented sentence information related to the determination, and outputs the information is described. Note that the future-oriented sentence is a sentence that corresponds to a future expression sentence and also corresponds to a purpose means sentence. The future expression sentence is a sentence including the future. The future expression sentence is, for example, a sentence that describes the future. The future-oriented sentence is, for example, a sentence that expresses the future in one sentence and includes at least one of the purpose and means related to the company. For example, “Future-oriented statements are as follows:“ In order to raise the position of the Group further to the top and to achieve stable sales, we will work to improve product competitiveness by strengthening our development system and expanding alliances. . " Here, it is determined that “for” is the purpose expression, “by planning” is the means expression, and “we will work” is the future expression sentence.

  A future expression sentence is usually a sentence that states what will be done in the future, or a sentence that states what is currently in progress. The future expression sentence is, for example, a sentence indicating the progress. Examples of sentences that indicate progress are: “Mr. Tanaka is reading a book in the room.” “Providing safety consulting aimed at achieving both safety and productivity.” The future expression sentence is, for example, a sentence indicating the remaining result relating to the result of something. Examples of statements that indicate the remaining results are, for example, “The window glass is broken.” “We have a production system in place to cope with disasters and unexpected situations.” The future expression sentence is, for example, a sentence indicating repetition. For example, “Mr. Tanaka is jogging every morning.” “A management meeting is held once a month as an organization that determines basic management policies and important matters and reviews and formulates issues related to the overall management cycle. It is held. " The future expression sentence is, for example, a sentence indicating that the effect is sustained with respect to the effect being maintained. For example, “This bridge was broken five years ago.” “We believe it is important to strengthen sales capabilities in the underlying domestic market, keeping in mind the new business portfolio. " The future expression sentence is, for example, a sentence indicating recording. For example, a sentence indicating the record states, “According to the book,“ I am saying this. ”“ Incorporating the needs of the times into corporate activities in response to changes in the environment, the corporate value and thus the common interests of shareholders are secured and improved. Is the top priority. ” The future expression sentence is, for example, a sentence indicating a completion phase (future). The completion phase (future) is a sentence indicating completion in the future. The statement indicating the completion phase (future) is, for example, “Aiming to develop and commercialize in-house services for Affectiva, Inc.'s emotion recognition AI invested through a corporate venture fund established in the United States.”

  Also, a sentence indicating completion, a sentence indicating counter-fact, a sentence indicating a simple state, and a sentence in a completion phase (current) indicating completion at present are not future expression sentences. A sentence indicating completion is, for example, “When he received a letter from him, he was already dead.” “With regard to new businesses that were unprofitable, there were factors such as worsening economic conditions, and many of them Has been withdrawn by the end of the previous consolidated fiscal year. " In addition, the sentence showing the counter-fact is, for example, “I was dead if he didn't help.” “In retrospect, the original subdivision redevelopment plan created on the desk of the government office was cut off. If so, we are convinced that the current existence of the company was not possible in the light of the economic situation from that time to the present day due to the decline in rental income and new investment. " In addition, a simple statement that states the state is, for example, “There is a high mountain on the north side of the school.” “In our group, the sales staff is called a field person.” Furthermore, the sentence indicating the completion phase (current) is, for example, “HDC stands for Hospitality, Delicious, and Cleanliness, and is a service improvement campaign for customers in our group.”

  The objective means sentence is a sentence having at least one of the objective expression and the means expression. The purpose expression may be an expression indicating the purpose or a character string including a term describing the purpose. The means expression may be an expression indicating the means or a character string including a term describing the means. An expression is a character string, such as a term, a word, or a phrase.

  The purpose means statement is, for example, “Aiming to build a business group with comprehensive competitiveness and work on reorganization and integration.” In this sentence, “Aim,” is a clue expression for purpose expression, and “I work” is a clue expression for means expression. The purpose means sentence is, for example, “We will increase the number of users by attracting customers”. In such a sentence, “by drawing” becomes a clue expression of means expression, and “I will advance” becomes a clue expression of purpose expression. In addition, the purpose means sentence is, for example, “Aiming to develop high-value-added products with superiority in the market by promoting aggressive research and development activities to meet customer needs”. In such a sentence, “To” is a clue expression of purpose expression, and “Promote” and “Aim” are clue expressions of means expression. In addition, the purpose means statement is, for example, “Strengthen the on-site support system by the head office and branch organization, work to reduce costs throughout the entire organization, improve the on-site capabilities by securing engineers and implementing appropriate allocation through the mobility of personnel. I will aim. " Note that the clue expression may be called a clue phrase.

  In the present embodiment, an information processing apparatus that performs a first filtering process that excludes a sentence that is clearly not a future expression sentence from two or more sentences will be described. A sentence that is not a future expression sentence is appropriately referred to as a non-future expression sentence. Further, in the present embodiment, the first filtering process may be performed as a pre-process for extracting the objective means sentence.

  Further, in the present embodiment, as a result of the first filtering process, a sentence that is clearly a future expression sentence or a sentence that is clearly not a future expression sentence is determined for the remaining sentences using a sentence end expression. An information processing apparatus that performs the second filtering process will be described.

  In the present embodiment, an information processing apparatus that determines a future expression sentence using a machine learning algorithm will be described.

  Also, in the present embodiment, an information processing apparatus that determines a target means sentence using a machine learning algorithm will be described.

  Furthermore, in the present embodiment, an information processing apparatus that performs statistical processing using one or more document attribute values associated with text information and acquires future-oriented text information will be described.

  FIG. 1 is a block diagram of an information processing apparatus A in the present embodiment. The information processing apparatus A includes a storage unit 1, a reception unit 2, a processing unit 3, and an output unit 4.

  The storage unit 1 includes a sentence information storage unit 11, a learning data storage unit 12, an exclusion condition storage unit 13, a sentence end condition storage unit 14, a first learner storage unit 15, and a second learner storage unit 16. The processing unit 3 includes a learning unit 31, an exclusion unit 32, a detection unit 33, and an acquisition unit 34. The detection unit 33 includes a future expression sentence detection means 331, a target means sentence detection means 332, and a selection means 333. The future expression sentence detection means 331 includes first machine learning means 3311. The objective means sentence detection means 332 includes second machine learning means 3321.

  Various information is stored in the storage unit 1. The various types of information include, for example, sentence information described later, learning data described later, exclusion conditions described later, sentence ending conditions described later, a first learner described later, and a second learner described later.

  The text information storage unit 11 stores one or more text information. The sentence information is information including one or two or more sentences. The text information is, for example, a business activity text describing business activities. The corporate activity text includes a securities report, a newsletter, a corporate website, and the like. However, the content of the text information does not matter. Moreover, although text information is a file, for example, the information stored in DB may be sufficient. The text information may be a text file or an HTML file. That is, the data structure of the text information, the description format, etc. are not questioned. Note that securities reports usually have a hierarchical structure. That is, the text information may be information on a document having a hierarchical structure or information on a document having a flat structure.

  The learning data storage unit 12 stores four or more learning data. The learning data is information serving as a basis for configuring the learning device. The learning device may be called a classifier. The four or more learning data includes two or more first learning data and two or more second learning data.

  The first learning data is learning data for configuring the first learning device. It is preferable that each of the two or more first learning data has one or more positive examples and one or more negative examples. A positive example of the first learning data includes a future expression sentence. The positive example of the first learning data may be information (flag) indicating a future expression sentence and a future expression sentence. The first learning data may be a future expression sentence. A negative example of the first learning data includes a non-future expression sentence. The negative example of the first learning data may be information (flag) indicating that it is not a future expression sentence and a non-future expression sentence. The negative example of the first learning data may be a non-future expression sentence.

  The second learning data is learning data for configuring the second learning device. Each of the two or more second learning data preferably has one or more positive examples and one or more negative examples. The positive example of the second learning data includes the objective means sentence. The positive example of the second learning data may be information (flag) indicating that it is an objective means sentence and an objective means sentence. The positive example of the second learning data may be a purpose means sentence. The negative example of the second learning data includes a sentence that is not the objective means sentence. The negative example of the second learning data may be information (flag) indicating that it is not the objective means sentence and a sentence that is not the objective means sentence. A negative example of the second learning data may be a sentence that is not a target means sentence. Note that a sentence that is not a purpose means sentence is appropriately referred to as a non-purpose means sentence.

  The exclusion condition storage unit 13 stores exclusion conditions. The exclusion condition is a condition for excluding sentences that do not correspond to future-oriented sentences. The exclusion condition is a condition for excluding a sentence that causes noise when, for example, a future-oriented sentence is extracted.

  The exclusion condition is, for example, a condition of one or more of a character number condition, an item condition, an attribute condition, and a character condition, or a combination of two or more conditions of a character number condition, an item condition, an attribute condition, and a character condition. .

  The character number condition is a condition related to the number of characters of a sentence. The character number condition is, for example, that the number of characters is less than or less than a threshold value. For example, the threshold is preferably 6, but may be 5 to 9 or the like.

  The item condition is an item in which a sentence appears, and is a condition relating to an item corresponding to the sentence. It may be said that the item condition is a condition related to a chapter or a section corresponding to the sentence. It may be said that the item condition is a condition related to the title corresponding to the sentence. The item condition is, for example, a sentence of a specific item or not a sentence of a specific item. The item condition is, for example, that it corresponds to the items of [issues to be addressed], [management policy, management environment, issues to be addressed, etc.] of the securities report. Items include headings, titles, chapters, sections, tags (for example, [issues to be addressed], [management policy, management environment, issues to be addressed, etc.]) and the like in a hierarchical document. It can be said that the item corresponding to the sentence is, for example, the item to which the sentence belongs.

  The attribute condition is a condition related to the attribute of the sentence. The attribute condition is that the sentence attribute is “title”, the sentence attribute is “text”, and the sentence attribute is “part”. The sentence attribute “title” means that the sentence is a heading in the text of the document. The sentence attribute “text” means that the sentence is a sentence in the text of the document. The sentence attribute “part” is a sentence part generated in the process of sentence structuring.

  The character condition is a condition related to a specific character included in the sentence. The character condition is, for example, a condition related to the character at the beginning of the sentence. The character condition is, for example, that there is a symbol at the beginning of the sentence and that there are no punctuation in the sentence. The symbols are, for example, “(1)”, “a)”, “(I)”, “(i)”, “•”, and the like.

For example, the exclusion condition corresponds to one of the following (1) to (4).
(1) The number of characters must be 6 or less. (2) The sentence must be in the item [issues to be addressed] or [management policy, management environment, issues to be addressed, etc.] (3) The attribute is the title & the symbol at the beginning of the sentence & there is no punctuation & the number of characters is 30 or less (4) The sentence attribute is part & no punctuation & the number of characters is 25 or less

  The sentence end condition storage unit 14 stores sentence end conditions. The sentence end condition is a condition related to the sentence end expression. The sentence end condition is one or more conditions among a condition for specifying a sentence corresponding to the future expression sentence and a condition for specifying a sentence not corresponding to the future expression sentence.

  The sentence end condition is, for example, whether or not it corresponds to the sentence end clue expression. The sentence ending conditions for specifying the sentence corresponding to the future expression sentence are, for example, “I will visit”, “I will do it”, “I will do it”, “I will do it”, and “I will do it”. That is, it is determined that a sentence having “I will come” or “I will do” at the end of the sentence is a sentence corresponding to the future expression sentence. For example, we will implement these measures, considering that it is indispensable to continuously make capital investments that take into account asset efficiency, including the introduction of new attractions in a timely manner. "We will focus on securing growth opportunities by developing new products in new markets, such as accelerating the development of immersion-related materials as materials for next-generation semiconductor manufacturing." It is judged. The sentence ending condition for specifying a sentence that does not correspond to the future expression sentence is, for example, a T-shape. A shape is a sentence whose end is "ta." That is, it is determined that the T-shaped sentence is a sentence that does not correspond to the future expression sentence. For example, the sentence “Agreed to develop as one of Asia's leading comprehensive energy, resource and material groups with international competitiveness through business integration and to contribute to the development of sustainable and vibrant economic and social development.” Etc. are judged to be non-future expressions.

  The first learner storage unit 15 stores the first learner. The first learner is information used when determining whether or not it is a future expression sentence by using two or more future expression positive example sentences, and is a learner created by a machine learning algorithm. Moreover, it is preferable that the first learning device is a learning device obtained by using one or more future expression negative example sentences in addition to one or more future expression negative example sentences. The future expression positive example sentence is a future expression sentence. The future expression negative example sentence is a non-future expression sentence. The machine learning algorithm will be described later.

  The first learning device is preferably information acquired by the learning unit 31. The first learner is, for example, a vector acquired using two or more future expression correct example sentences, sentence ending clue expression information related to a predetermined sentence ending clue expression, and frequent terms that are predetermined frequent terms. Frequently used term information, adverb information about adverb terms in sentences, pair information about two-sentence pairs at the end of a sentence, part of speech information about morpheme at the end of a sentence, punctuation information about punctuation included in a sentence, sentence attributes about sentence attributes It is preferable that the learning device be learned and acquired by a machine learning algorithm using two or more vectors having one or more pieces of information as elements. The first learning device may be, for example, a learning device that is learned and acquired by a machine learning algorithm using two or more vectors having one or more pieces of information of morpheme unigrams and morpheme bigrams as elements. .

  Sentence clue expression information, frequent term information, adverb information, pair information, part of speech information, punctuation information, and sentence attribute information will be described later.

  The second learner storage unit 16 stores the second learner. The second learner is information used when determining whether or not the sentence is a target means sentence using two or more target means positive example sentences, and is a learner created by a machine learning algorithm. Moreover, it is preferable that the second learning device is a learning device obtained by using one or more target means negative example sentences in addition to one or more target means positive example sentences. The target means positive example sentence is a target means sentence. The target means negative example sentence is a non-purpose means sentence.

  The second learning device is preferably information acquired by the learning unit 31. The second learner is, for example, a vector acquired using two or more objective means positive example sentences, and the part-of-speech part-of-speech information about the part-of-speech information about the part-of-speech of the word before the punctuation point, Information of one or more of information, punctuation information about punctuation included in the sentence, sentence attribute information about the attribute of the sentence, and first determination result information about whether or not it is a future expression sentence acquired by the first machine learning means 3311 It is preferable that the learning device is obtained by learning with an algorithm of machine learning using two or more vectors having elements as. The second learning device may be, for example, a learning device that has been learned and acquired by a machine learning algorithm using two or more vectors whose elements are one or more of morpheme unigrams and morpheme bigrams. .

  Part-of-speech information before punctuation, part-of-speech part-of-speech information, punctuation information, sentence attribute information, and first determination result information will be described later.

  The accepting unit 2 accepts various instructions and information. The various instructions and information are, for example, a start instruction that is an instruction to start an operation, a first learning instruction that is an instruction that constitutes a first learner, a second learning instruction that is an instruction that constitutes a second learner, The sentence information, the learning data, the exclusion condition, the sentence end condition, the first learning device, and the second learning device. Here, reception means reception of information input from an input device such as a keyboard, mouse, touch panel, reception of information transmitted via a wired or wireless communication line, recording on an optical disk, magnetic disk, semiconductor memory, or the like. It is a concept including reception of information read from a medium. Any means such as a touch panel, a keyboard, a mouse, or a menu screen may be used for inputting various instructions and information. The accepting unit 2 can be realized by a device driver for input means such as a touch panel and a keyboard, control software for a menu screen, and the like.

  The processing unit 3 performs various processes. The various processes are, for example, processes performed by the learning unit 31, the exclusion unit 32, the detection unit 33, the acquisition unit 34, and the like.

  For example, the processing unit 3 may acquire various clue expressions from one or more pieces of text information by bootstrap. The various clue expressions are a character string for extracting a future expression sentence, a character string for determining that it is not a future expression sentence, a character string indicating a purpose expression, a character string indicating means expression, and the like. Moreover, since bootstrap is a well-known technique, detailed description is abbreviate | omitted. Regarding bootstrap, see, for example, the document “[Hiroyuki Sakai, Shigeru Masuyama,“ Cause Information Extraction from Financial Articles Concerning Business Performance ”, IEICE Trans. Information and Systems, ED, 4, pp. 959-968, 2008.] about.

  The learning unit 31 uses the two or more first learning data items stored in the learning data storage unit 12 to acquire a first learning device using a machine learning algorithm. The machine learning algorithm is, for example, SVM, deep learning, decision tree, random forest, etc., but other machine learning algorithms may be used.

  The learning unit 31 stores, for example, machine representation functions (for example, TinySVM, fastText, various RandomForest functions, etc.) as future input correct example sentences (for example, future expression correct example sentences themselves, future expression correct example sentences). File or its file pointer) and information to be output (a flag indicating that it is a positive example, for example, “1”) is given as two or more pairs as arguments, and the function is executed. Get the first learner.

  For example, the learning unit 31 wants a machine learning function to output a future expression correct example sentence (for example, a future expression correct example sentence itself, a file in which the future expression correct example sentence is stored, or a file pointer thereof). 2 or more sets of information (a flag indicating that it is a positive example, for example, “1”) and an input future expression negative example sentence (for example, a future expression negative example sentence itself, a future expression negative example sentence are stored. File or its file pointer, etc.) and the information to be output (a flag indicating that it is a negative example, for example, “0”) is given as one or more sets of arguments to obtain the first learner To do.

  The learning unit 31 uses the two or more second learning data in the learning data storage unit 12 to acquire a second learning device by a machine learning algorithm.

  For example, the learning unit 31 wants a machine learning function to output the target means positive example sentence (for example, the target means positive example sentence itself, a file storing the target means correct example sentence, or a file pointer thereof). Two or more sets of information (a flag indicating that it is a positive example, for example, “1”) is given as an argument, and a second learner is acquired.

  The learning unit 31 includes, for example, two or more sets of a target example positive example sentence that is an input to a machine learning function and information to be output (a flag indicating a positive example, for example, “1”). , And the target means negative example sentence that is input and the information to be output (a flag indicating that it is a negative example, for example, “0”) is given as one or more sets as arguments, and the second learner is acquired. To do.

  The exclusion unit 32 determines whether the sentence satisfies the exclusion condition. The exclusion unit 32 performs a first filtering process on each sentence to be examined. For example, the exclusion unit 32 determines whether one or more sentences included in one or more pieces of sentence information satisfy the exclusion condition, and excludes sentences that satisfy the exclusion condition. The processing to be excluded here is, for example, processing that deletes the corresponding sentence, processing that adds a flag in association with the corresponding sentence, and processing that makes it possible to distinguish a sentence that is not excluded from an excluded sentence. Anything is fine. The exclusion unit 32 is not an essential element. That is, the processing of the detection unit 33 to be described later may be executed for all sentences to be examined.

  For example, the excluding unit 32 operates as follows. First, the exclusion unit 32 obtains information on one or more of the number of characters, the item name to which the sentence belongs, the first attribute of the sentence, the second attribute of the sentence, the character at the beginning of the sentence, and the punctuation from the sentence to be examined. To do. The first attribute of the sentence is information indicating whether the sentence is a title. The second attribute of the sentence is information indicating whether the sentence is a part. A technique for acquiring the number of characters, the first letter, a punctuation point, and the like from a sentence is a well-known technique.

  The exclusion unit 32 is, for example, the closest position of a sentence, and a heading that appears before the sentence (for example, a sentence whose sentence starts with “numerical value” + “.” (A sentence in a chapter or section number line), tag (Sentence enclosed by “[” and “]”)), and the item name to which the sentence belongs is acquired. Such processing can be performed by pattern matching, machine learning, or the like.

  For example, the excluding unit 32 determines whether or not the sentence is a title based on whether or not a predetermined character string (for example, the beginning of the sentence is “numerical value” + “.”) Exists before the sentence. Further, the excluding unit 32 determines whether the sentence is a title based on, for example, whether the sentence is a sentence surrounded by tags (“[” and “]”). In addition, the process which determines whether a sentence is a title does not ask | require. With this determination process, the exclusion unit 32 can acquire the first attribute of the sentence.

  Further, the excluding unit 32 determines that, for example, a part whose sentence attribute cannot be determined is a part that is not a predetermined attribute (for example, “title”, “text”, and the like). In addition, the process which determines whether a sentence is a part is not ask | required. With this determination process, the exclusion unit 32 can acquire the second attribute of the sentence.

  Then, the exclusion unit 32 stores the sentence using one or more information of the acquired number of characters, the item name to which the sentence belongs, the first attribute of the sentence, the second attribute of the sentence, the character at the beginning of the sentence, and the punctuation mark. It is determined whether or not the exclusion condition of part 1 is met.

For example, the excluding unit 32 excludes sentences corresponding to any of the following excluding conditions (1) to (4).
(1) The number of characters must be 6 or less. (2) The sentence must be in the item [issues to be addressed] or [management policy, management environment, issues to be addressed, etc.] (3) The attribute is the title & the symbol at the beginning of the sentence & there is no punctuation & the number of characters is 30 or less (4) The sentence attribute is part & no punctuation & the number of characters is 25 or less

  The detection unit 33 determines whether the sentence is a future-oriented sentence. It is preferable that the detection unit 33 detects a future-oriented sentence from one or more sentences included in the one or more pieces of sentence information stored in the sentence information storage unit 11.

  The detection unit 33 is, for example, one or more sentences included in one or more pieces of sentence information, and is a future expression sentence and a target means sentence from one or more sentences excluding the sentence excluded by the exclusion unit 32. Detect oriented sentences.

  However, when the exclusion unit 32 does not exist, the detection unit 33 examines all the sentences to be examined and determines whether or not the sentence is a future-oriented sentence.

  For example, the detection unit 33 may determine that the sentence matching the sentence end condition is a future expression sentence. For example, the detection unit 33 may determine that the sentence that matches the sentence end condition is not a future expression sentence.

  For example, the detection unit 33 determines whether the sentence is a future-oriented sentence using a clue expression. For example, if a predetermined character string (cue expression) such as “I will come” or “I will come” is present in the sentence, the detection unit 33 determines that the sentence is a future expression sentence. For example, if a predetermined character string such as “for” is present in the sentence, the detection unit 33 determines that the sentence is a purpose expression sentence. For example, if a predetermined character string such as “by drawing” is present in the sentence, the detection unit 33 determines that the sentence is a means expression. In such a case, a predetermined character string is stored in the storage unit 1. Then, the detection unit 33, for example, matches the condition “future expression sentence AND (object expression sentence OR statement of means expression)” (future expression sentence and statement of purpose expression or means expression). ), The sentence is determined to be a future-oriented sentence, and if the above condition is not met, it is determined that the sentence is not a future-oriented sentence.

  The detection unit 33 uses the future expression sentence detection unit 331 described later, the target means sentence detection unit 332 described later, and the selection unit 333 described later to determine whether the sentence to be examined is a future-oriented sentence. Is preferred but is not constrained by such methods. As described above, it may be determined whether the sentence is a future-oriented sentence using only the clue expression.

  The future expression sentence detection unit 331 determines whether or not the examination target sentence is a future expression sentence.

  The future expression sentence detection unit 331 may determine whether or not the sentence to be examined is a future expression sentence using a clue expression indicating the future expression. That is, for example, the future expression sentence detection unit 331 determines that a sentence including a clue expression indicating the future expression is a future expression sentence, and determines that a sentence not including the clue expression indicating the future expression is not a future expression sentence. Also good. Further, it is possible to determine whether or not the sentence to be examined is a future expression sentence by using a clue expression indicating that it is not a future expression. That is, the future expression sentence detection unit 331 may determine that a sentence including a clue expression indicating that it is not a future expression is not a future expression sentence. In addition, clue expressions indicating the future expression are, for example, “I will visit”, “I will continue”, “I will do”, “I will do”. Further, the clue expression indicating that it is not the future expression is, for example, “ta.” Further, a clue expression indicating the future expression and a clue expression indicating that it is not the future expression are stored in the storage unit 1.

  The future expression sentence detection unit 331 may determine whether or not the sentence to be examined is a future expression sentence by using a clue expression indicating the future expression without using a machine learning algorithm described later. Further, the future expression sentence detection means 331 determines whether the sentence to be examined is a future expression sentence by using a clue expression indicating that it is not the future expression without using a machine learning algorithm described later. May be.

  The future expression sentence detection means 331 uses the clue expression indicating the future expression to determine whether the sentence to be inspected is a future expression sentence, and the first machine learning means 3311 described later uses a machine learning algorithm. It is preferable to determine whether or not the sentence to be examined is a future expression sentence. Further, the future expression sentence detection unit 331 determines whether the sentence to be inspected is a future expression sentence by using a clue expression indicating that it is not the future expression, and the first machine learning means 3311 described later It is preferable to determine whether the sentence to be inspected is a future expression sentence by using a machine learning algorithm. In such a case, using the clue expression indicating the future expression, a process for determining whether the sentence to be inspected is a future expression sentence, or using the clue expression indicating that the sentence to be inspected is not the future expression, The process for determining whether or not it is a future expression sentence is a second filtering process.

  The first machine learning means 3311 constituting the future expression sentence detection means 331, for example, uses the first learner to determine whether or not the sentence to be examined is a future expression sentence by a machine learning algorithm. decide. The first machine learning means 3311 is a future expression sentence by a machine learning algorithm for the remaining sentence that has not been determined whether or not it is a future expression sentence as a result of the second filtering process. It is preferable to determine whether or not. However, the first machine learning means 3311 may determine whether or not all sentences to be examined are future expression sentences by a machine learning algorithm. Note that the process for determining whether or not it is a future expression sentence may be a process that distinguishes a future expression sentence from a non-future expression sentence. For example, information indicating whether or not it is a future expression sentence is acquired. For example, processing, processing for deleting non-future expressions, and storing future expressions and non-future expressions in different storage areas.

  The first machine learning unit 3311 applies, for example, a sentence to be examined to the first learning device, and acquires information indicating whether or not the sentence is a future expression sentence. When the first machine learning means 3311 gives, for example, a machine learning function (for example, TinySVM, fastText, various RandomForest functions, etc.) as arguments, the first learner and one or more sentences to be inspected are given as 1 Information predicted for each of the above sentences (information indicating whether each sentence is a future expression sentence) is obtained.

  The first machine learning means 3311 determines, for example, one or more of sentence end clue expression information, frequent term information, adverb information, pair information, phrase information, part of speech information, phrase information, and sentence attribute information from the sentence to be examined. And a vector having the one or more pieces of information as elements. Then, the first machine learning means 3311 applies the vector to the first learner, and acquires information indicating whether or not it is a future expression sentence by a machine learning algorithm.

  More specifically, the first machine learning unit 3311 gives a learner and the vector as arguments to a machine learning function, executes the function, and provides information (information indicating whether it is a future expression sentence). get. The first machine learning means 3311 adds one or more of morpheme unigram and morpheme bigram from the obtained morpheme in addition to the above one or more information from the sentence to be examined. It is preferable to construct a vector having such information as an element. In this case, the vector is a vector having, for example, morpheme unigram, morpheme bigram, sentence end clue expression information, frequent term information, adverb information, pair information, phrase information, part of speech information, phrase information, and sentence attribute information as elements.

  A morpheme unigram is one morpheme. The first machine learning means 3311 performs morpheme analysis on the sentence and acquires each morpheme. In addition, morphological analysis is performed using MeCab (URL: https://www.mlab.im.dendai.ac.jp/~yamada/ir/MorphologicalAnalyzer/MeCab.html), Chasen (URL: http: // chasen. naist.jp/hiki/ChaSen/?%C3%E3%E4%A5%A4%CE%C7%DB%C9%DB etc.), etc.

  A morpheme bigram is two consecutive morphemes in a sentence. The first machine learning unit 3311 performs morphological analysis on the sentence and acquires two consecutive morphemes as one piece of information.

  The sentence end clue expression information is information indicating whether or not the sentence end clue expression matches a specific clue expression at the end of the sentence. For example, the first machine learning unit 3311 detects whether or not the end of a sentence matches any one of the clue expressions in the storage unit 1, and obtains different values depending on whether or not they match. For example, the first machine learning unit 3311 acquires “1” if they match, and acquires “0” if they do not match. In such a case, the sentence ending clue expression information has one element. Further, the first machine learning unit 3311 may acquire sentence ending clue expression information including, for example, each piece of information as to whether or not it matches each specific clue expression at the end of the sentence. In such a case, the sentence ending clue expression information has two or more elements.

  The specific clue expression at the end of the sentence is, for example, FIG. Such clue expression is, for example, a character string acquired by the bootstrap described above.

  More specifically, the processing unit 3 visits the seed expression ““. "I will do it." "," Do it. "," Do it. “” And α = 0.45, and bootstrap is performed to obtain the clue expression shown in FIG. Then, the first machine learning means 3311 acquires the output excluding the expression that does not end with a punctuation and the expression that the sentence end is “ta” as the feature of the sentence end clue expression. Next, the first machine learning unit 3311 parses the sentence, determines whether the sentence end expression matches each element of the feature (for example, the sentence end clue expression in FIG. 2), and acquires sentence end clue expression information.

  The frequently used term information is information on frequently used terms that are predetermined frequently used terms. For example, the frequent term information takes different values depending on whether it matches any of the frequent terms in the storage unit 1 or not, for example, “1” if they match, and “0” if they do not match. Is. In such a case, the frequently used term information is information of one element. For example, when there are two or more frequently used terms, the element corresponding to the frequently used term is “1”, and when each frequently used term does not exist, the element corresponding to the frequently used term is “0”. is there. In such a case, the frequently used term information is information of two or more elements.

  For information on how to obtain frequently used term information, see, for example, the document “[Hiroyuki Sakai, Shigeru Masuyama,“ Cause Information Extraction from Financial Articles Concerning Business Performance ”, IEICE Trans. Information and Systems, ED, 4, pp.959-968. , 2008.], it goes without saying that it is not bound by such a method.

  For example, the processing unit 3 acquires a common frequent expression (frequently used term) output when acquiring a sentence end cue expression as a feature of the common frequent expression. Then, for example, the first machine learning unit 3311 parses the sentence, determines whether the phrase expression related to the sentence end phrase matches each element of the feature (common frequent expression), and acquires frequent term information.

  The adverb information is information related to the term of the adverb included in the sentence. The processing unit 3 performs a morphological analysis on each sentence of the learning data, and accumulates morphemes whose adverbs are adverbs in the storage unit 1 except for duplication. And the process part 3 makes the acquired morpheme the feature of an adverb. Then, the first machine learning means 3311 determines whether or not each feature element (adverb morpheme) is included in the sentence to be examined, and acquires adverb information as a determination result. The first machine learning means 3311 may acquire adverb information indicating whether or not a morpheme of any adverb is included in the sentence to be examined. In such a case, the adverb information is information of one element. Further, the first machine learning means 3311 may acquire adverb information indicating whether or not the sentence to be examined includes morphemes of two or more adverbs. In such a case, the adverb information is information of two or more elements.

  The pair information is information related to a pair of expressions in the last two phrases. The pair information is information on a pair of expressions of two phrases at the end of the sentence. For example, the processing unit 3 parses each sentence of the learning data, and accumulates pairs of sentence end 2 phrase expressions in the storage unit 1 excluding duplicates. Then, the processing unit 3 sets the acquired phrase expression pair as the feature of the sentence end two phrase pair. The processing unit 3 parses each sentence of the learning data, creates an expression (character string) of two clauses at the end of the sentence, determines whether or not each element of the feature (a pair of clauses) matches, and is a determination result. Get pair information. The first machine learning means 3311 may acquire pair information indicating whether or not the sentence to be examined includes a pair of expressions of two or more sentence end two clauses. In such a case, the pair information is information of two or more elements. Further, the first machine learning means 3311 may acquire pair information indicating whether or not the sentence to be examined includes a pair of expressions of any one sentence end two clauses. In such a case, the pair information is information of one element.

  The part of speech information is information related to the part of speech of the morpheme at the end of the sentence. For example, the processing unit 3 performs morphological analysis on each sentence of the learning data, and accumulates the part of speech of the morpheme at the end of the sentence (one before the punctuation mark) in the storage unit 1 excluding duplication. And the process part 3 makes the acquired part of speech the feature of the part of speech of the sentence end, for example. The first machine learning means 3311 performs a morphological analysis on the sentence to be inspected, determines whether the part of speech of the morpheme at the end of the sentence (one before the punctuation mark) matches each element (part of speech) of the feature, and the part of speech that is the determination result Get information. The part of speech information is information indicating whether or not the part of speech of the morpheme at the end of the sentence matches each element of two or more features, and may be information of two or more elements. The part of speech information is information indicating whether the part of speech of the morpheme at the end of the sentence matches an element of any feature, and may be information of one element.

  Further, the phrase information is information regarding the phrase included in the sentence. The first machine learning unit 3311 acquires, for example, phrase information that is information indicating whether or not the sentence to be examined includes a phrase. The first machine learning unit 3311 obtains, for example, punctuation information indicating the number including punctuation in the sentence to be examined.

  The sentence attribute information is information regarding sentence attributes. The sentence attribute information is, for example, information indicating whether the sentence exists in the body text, exists in the title, is a component, or is a bulleted sentence. The sentence attribute information may be information specifying whether the sentence exists in the body text, in the title, whether it is a part, or whether it is a bulleted sentence. Information is information of one element. The sentence attribute information includes each of two or more items of whether the sentence exists in the text, whether it exists in the title, whether it is a part, or whether it is a bulleted sentence. In this case, the sentence attribute information is information of two or more elements.

  The first machine learning unit 3311 acquires an item to which the sentence belongs, and acquires sentence attribute information using the item.

  As described above, for example, the first machine learning unit 3311 configures a vector having the number of dimensions shown in FIG. In FIG. 3, the unigram is the number of dimensions of the morpheme unigram, the bigram is the number of dimensions of the morpheme bigram, the clue expression is the number of dimensions of the end-of-sentence clue expression information, the common frequent expression is the number of dimensions of frequent term information, the adverb is the number of dimensions of adverb information, The sentence end 2 phrase pair is the number of dimensions of the pair information, the part of speech of the sentence is the number of dimensions of the part of speech information, the punctuation is the number of dimensions of the punctuation information, the sentence attribute is the number of dimensions of the sentence attribute information, and the total is the number of dimensions of the vector.

  The objective means sentence detection means 332 determines whether the sentence to be examined is an objective means sentence.

  The second machine learning unit 3321 constituting the target unit sentence detection unit 332 uses, for example, a second learner to perform an objective on one or more sentences included in one or more pieces of sentence information by a machine learning algorithm. Determine whether it is an instrumental statement.

  However, the objective means sentence detection means 332 determines whether or not the sentence to be examined is the objective means sentence using a clue expression indicating one or more types of expressions of the objective expression and the means expression. Also good. In other words, the purpose means sentence detection means 332 determines that, for example, a sentence including a clue expression indicating a purpose expression or means expression is a purpose means sentence, and a sentence not including the purpose expression and the clue expression indicating the means expression is a purpose means. You may judge that it is not a sentence.

  For example, the second machine learning unit 3321 applies the sentence to be examined to the second learning device, and acquires information indicating whether or not the sentence is the target unit sentence. When the second machine learning unit 3321 gives, for example, a machine learning function (for example, TinySVM, fastText, various RandomForest functions, etc.) as arguments, the second learner and the sentence to be examined are predicted information (purpose Information indicating whether or not it is a means sentence) is obtained.

  For example, the second machine learning unit 3321 removes, from the sentence to be inspected, the sentence for which the target unit sentence detection unit 332 uses the clue expression to determine whether the target unit sentence is the target unit sentence. Applying to the second learning device, information indicating whether or not the sentence is the objective means sentence may be acquired. In such a case, the process using the clue expression of the target means sentence detection means 332 is a third filtering process.

  For example, the second machine learning unit 3321 includes, from among one or more sentences included in one or more pieces of sentence information, part-of-speech information before punctuation, part-of-speech information before punctuation, punctuation information, sentence attribute information, and first determination result information. Get one or more pieces of information, construct a vector with one or more pieces of information as elements, apply the vector to the second learner, and get information that indicates whether it is a target means sentence by machine learning algorithm To do.

  The second machine learning unit 3321 performs morphological analysis on the sentence to be examined in addition to the one or more pieces of information from the sentence to be examined. From the obtained morpheme, one or more types of morpheme unigram and morpheme bigram are obtained. It is preferable to construct a vector having such information as an element. In this case, the vector is a vector having, for example, morpheme unigrams, morpheme bigrams, part-of-speech information before reading, part-of-speech information, punctuation information, sentence attribute information, and first determination result information as elements.

  The part-of-speech information before punctuation is information related to the part of speech of the morpheme immediately before the punctuation. When a sentence (,) is included in each sentence of the learning data, the processing unit 3 performs a morphological analysis on the sentence, and accumulates the part of speech of the morpheme immediately preceding the reading point in the storage unit 1 without duplication. Then, the processing unit 3 sets the acquired part of speech as the part of speech before the reading. Then, the second machine learning means 3321 performs a morphological analysis on the sentence to be examined, determines whether the part of speech of one morpheme before the reading is coincident with each element (part of speech) of the feature, and before the reading as the determination result Get part of speech information. Note that the second machine learning means 3321 acquires an element indicating “not included” when the sentence does not include a punctuation mark, for example. The second machine learning means 3321 determines, for example, whether the part of speech of one morpheme before the reading matches with each element (part of speech) of two or more features, and information for each part of speech (the morpheme before the reading of the part of speech) Whether or not there exists) may be acquired. In such a case, the part-of-speech information before reading has two or more elements. Further, the second machine learning means 3321 determines, for example, whether the part of speech of one morpheme before the punctuation coincides with any element (part of speech) of the feature, and obtains the part of speech information before punctuation consisting of one element. You may do it.

  The part-of-speech information before the punctuation is information related to the part-of-speech of the morpheme immediately before the punctuation. The processing unit 3 performs morphological analysis on each sentence of the learning data, and accumulates the part of speech of the morpheme at the end of the sentence (one before the punctuation mark) in the storage unit 1 without duplication. Then, the processing unit 3 uses the acquired part of speech as the feature of the part of speech at the end of the sentence. Then, the second machine learning means 3321 performs a morphological analysis on the sentence to be examined, determines whether the morpheme of the morpheme at the end of the sentence (one before the punctuation mark) matches each element (part of speech) of the feature, Acquire part-of-speech part-of-speech information. The second machine learning means 3321 determines, for example, whether or not the part of speech of one morpheme before the punctuation matches each element (part of speech) of two or more features, and information for each part of speech (morpheme before the punctuation of the part of speech. Whether or not there exists) may be acquired. In such a case, the pre-punctuation part-of-speech information has two or more elements. Also, the second machine learning means 3321 determines, for example, whether the part of speech of one morpheme before the punctuation matches any element (part of speech) of the feature, and acquires pre-reading part of speech information consisting of one element. You may do it.

  The first determination result information is a result of the detection unit 33 (for example, the first machine learning unit 3311) determining whether or not the sentence is a future expression sentence. The first determination result information may be information of two elements indicating, for example, whether the sentence is a future expression sentence or not, and whether the sentence is not a future expression sentence. For example, the first determination result information may be “1” when the sentence is a future expression sentence, and may be “0” when the sentence is not a future expression sentence. When the first determination result information is used, it is assumed that it is determined whether the sentence is a future expression sentence before determining whether the sentence is a target means sentence.

  As described above, for example, the second machine learning unit 3321 configures a vector having the number of dimensions shown in FIG. In FIG. 4, the unigram is the number of dimensions of the morpheme unigram, the bigram is the number of dimensions of the morpheme bigram, the number of parts of speech part-of-speech information before the punctuation mark, the part of speech before the punctuation point is the dimension number of the part-of-speech part-of-speech information, The number of dimensions of the phrase information, the attribute of the sentence is the number of dimensions of the sentence attribute information, the result of the SVM first stage is the number of dimensions of the first determination result information, and the total is the number of dimensions of the vector.

  The order of the first determination for determining whether or not the detection unit 33 is a future expression sentence and the second determination for determining whether or not it is a purpose means sentence are not limited. The second determination may be performed after the first determination, or the first determination may be performed after the second determination. Further, the first determination and the second determination may be performed in parallel. Moreover, the detection part 33 may perform the process of a 2nd determination only to the sentence determined to correspond to a future expression sentence as a result of a 1st determination. Moreover, the detection part 33 may perform the process of a 1st determination only to the sentence determined to correspond to an objective means sentence as a result of a 2nd determination.

  The selection unit 333 selects a future-oriented sentence that is a sentence that the future expression sentence detection means 331 determines to be a future expression sentence and the purpose means sentence detection means 332 determines to be a target means sentence. The sentence selection may be acquisition of a sentence, addition of a flag in association with the sentence, deletion of a sentence other than a future-oriented sentence, or the like. That is, the selection unit 333 may distinguish between future-oriented sentences and non-future-oriented sentences.

  Further, when the first determination and the second determination are sequentially performed, the result of the second determination or the result of the first determination is a result of whether or not the sentence is a future-oriented sentence. In such a case, the selection means 333 is not necessary.

  The acquisition unit 34 acquires future-oriented sentence information that is information related to the determination result in the detection unit 33. The acquisition unit 34 acquires future-oriented text information that is information related to the future-oriented text detected by the detection unit 33.

  The acquisition unit 34 acquires, for example, future-oriented sentence information that is information on the future-oriented sentence selected by the selection unit 333.

  The acquisition unit 34 performs, for example, statistical processing using one or more future-oriented sentences detected by the detection unit 33, and acquires future-oriented sentence information that is a statistical processing result. The future-oriented sentence information is, for example, the number of future-oriented sentences in the sentence information and the ratio of the future-oriented sentences in the sentence information.

  The acquisition unit 34 acquires, for example, future-oriented sentence information that is a statistical processing result using a sentence attribute value. The text attribute value is, for example, a company identifier that identifies a company or a business type identifier that identifies a business type.

  For example, the acquisition unit 34 acquires the future-oriented sentence information that is the statistical processing result by using the company identifier or the business type identifier. In this case, the future-oriented sentence information is, for example, the number of future-oriented sentences in the sentence information for each company and the ratio of the future-oriented sentences in the sentence information. Further, the future-oriented sentence information is, for example, the number of future-oriented sentences in the sentence information for each type of business and the ratio of the future-oriented sentences in the sentence information. Further, the future-oriented sentence information is, for example, the number of future-oriented sentences in the sentence information compared to other companies in the industry to which the company belongs, and the ratio of the future-oriented sentences in the sentence information.

  Further, the acquisition unit 34 may calculate the score using, for example, information on the number of future-oriented sentences in the sentence information or the ratio of future-oriented sentences in the sentence information. Such a score is calculated by an increasing function using the number of future-oriented sentences or the ratio of future-oriented sentences as a parameter. It is assumed that the information on the arithmetic expression of the increase function is stored in the storage unit 1.

  Further, the acquisition unit 34 may calculate a score for each company identifier, for example. Moreover, the acquisition part 34 may calculate the relative score compared with another company for every company identifier, for example. The acquisition unit 34 may calculate a relative score (for example, a deviation value) in one industry for each company identifier, for example.

  The output unit 4 outputs the future-oriented sentence information acquired by the acquisition unit 34. Here, output refers to display on a display, projection using a projector, printing with a printer, sound output, transmission to an external device, storage in a recording medium, and output to other processing devices or other programs. It is a concept that includes delivery of processing results.

  The storage unit 1, the sentence information storage unit 11, the learning data storage unit 12, the exclusion condition storage unit 13, the sentence end condition storage unit 14, the first learning device storage unit 15, and the second learning device storage unit 16 are nonvolatile recordings. A medium is preferred, but a volatile recording medium can also be realized.

  The process in which information is stored in the storage unit 1 or the like is not limited. For example, information may be stored in the storage unit 1 or the like via a recording medium, information transmitted via a communication line or the like may be stored in the storage unit 1 or the like, Alternatively, information input via the input device may be stored in the storage unit 1 or the like.

  Processing unit 3, learning unit 31, exclusion unit 32, detection unit 33, acquisition unit 34, future expression sentence detection unit 331, purpose unit sentence detection unit 332, selection unit 333, first machine learning unit 3311, and second machine learning The means 3321 can be usually realized by an MPU, a memory, or the like. The processing procedure of the processing unit 3 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The output unit 4 may be considered as including or not including an output device such as a display or a speaker. The output unit 4 can be realized by driver software for an output device or driver software for an output device and an output device.

  Next, an operation example of the information processing apparatus A will be described using the flowchart of FIG.

  (Step S501) The accepting unit 2 determines whether a start instruction has been accepted. When the start instruction is accepted, the process goes to step S502. When the start instruction is not accepted, the process goes to step S513.

  (Step S502) The processing unit 3 substitutes 1 for the counter i.

  (Step S503) The processing unit 3 determines whether or not the i-th sentence information exists in the sentence information storage unit 11. If i-th sentence information exists, it will go to step S504, and if it does not exist, it will go to step S511.

  (Step S504) The detection unit 33 acquires the i-th sentence information from the sentence information storage unit 11.

  (Step S505) The detection unit 33 substitutes 1 for a counter j.

  (Step S506) The detection unit 33 determines whether or not the j-th sentence exists in the sentence information acquired in step S504. If the jth sentence exists, go to step S507, and if the jth sentence does not exist, go to step S510. Note that a technique for detecting sentence breaks and acquiring sentences from sentence information is a well-known technique, and a detailed description thereof will be omitted. For example, the detection unit 33 extracts a sentence using clues as clues. Further, the detection unit 33 extracts a sentence using the return code as a clue.

  (Step S507) The detection unit 33 acquires the j-th sentence from the sentence information acquired in Step S504.

  (Step S508) The detection unit 33 performs detection processing on the sentence acquired in step S507. The detection process is a process for determining whether or not the sentence is a future-oriented sentence. The detection process will be described with reference to the flowchart of FIG.

  (Step S509) The detector 33 increments the counter j by 1. The process returns to step S506.

  (Step S510) The processing unit 3 increments the counter i by one. The process returns to step S503.

  (Step S511) The acquisition unit 34 acquires future-oriented sentence information. Such acquisition processing will be described with reference to the flowchart of FIG.

  (Step S512) The output unit 4 outputs the future-oriented sentence information acquired in Step S511. The process returns to step S501.

  (Step S513) The reception unit 2 determines whether or not a first learning instruction has been received. When the first learning instruction is accepted, the process goes to step S514. When the first learning instruction is not accepted, the process goes to step S515.

  (Step S514) The learning unit 31 performs a first learning process, acquires a first learner, and accumulates the first learner in the first learner storage unit 15. The first learning process will be described using the flowchart of FIG.

  (Step S515) The reception unit 2 determines whether a second learning instruction has been received. When the second learning instruction is accepted, the process goes to step S516, and when the second learning instruction is not accepted, the process returns to step S501.

  (Step S516) The learning unit 31 performs a second learning process, acquires a second learning device, and accumulates the second learning device in the second learning device storage unit 16. A 2nd learning process is demonstrated using the flowchart of FIG.

  In the flowchart of FIG. 5, the process ends when the power is turned off or the process ends.

  Next, an example of the detection process in step S508 will be described using the flowchart of FIG.

  (Step S601) The exclusion unit 32 performs an exclusion process. Exclusion processing will be described with reference to the flowchart of FIG.

  (Step S602) The detection unit 33 determines whether the sentence to be processed is a sentence that is excluded as a result of the exclusion process in Step S601. If the sentence is excluded, the process goes to step S608. If the sentence is not excluded, the process goes to step S603.

  (Step S603) The detection unit 33 performs future expression sentence inspection. The future expression sentence inspection is a process for inspecting whether a sentence is a future expression sentence or a non-future expression sentence. The future expression sentence inspection will be described with reference to the flowchart of FIG.

  (Step S604) The detection unit 33 performs an objective means sentence inspection. The purpose means sentence check is a process for checking whether a sentence is a purpose means sentence or a non-purpose means. The objective means inspection will be described with reference to the flowchart of FIG.

  (Step S605) The detection unit 33 determines whether the sentence is a “future expression sentence” and a “target means sentence” as a result of the inspections in steps S603 and S604. If this condition is met, the process goes to step S606, and if not, the process goes to step S607.

  (Step S606) The detection unit 33 substitutes information (for example, “1”) indicating “future-oriented sentence” into the variable “judgment result”. Return to upper process.

  (Step S607) The detection unit 33 substitutes information (for example, “0”) indicating “not a future-oriented sentence” into the variable “judgment result”. Return to upper process.

  (Step S608) The detection unit 33 substitutes information (for example, “1”) indicating “not a future-oriented sentence” into the variable “judgment result”. Return to upper process.

  In the flowchart of FIG. 6, the exclusion process may not exist.

  Next, an example of the exclusion process in step S601 will be described using the flowchart of FIG.

  (Step S701) The exclusion unit 32 acquires the number of characters of the sentence to be examined.

  (Step S702) The exclusion unit 32 acquires the item name to which the sentence belongs.

  (Step S703) The exclusion unit 32 acquires information indicating whether or not the sentence is a title (information indicating whether or not the sentence is a sentence in the title).

  (Step S704) The exclusion unit 32 acquires information indicating whether or not the sentence is a part.

  (Step S705) The exclusion unit 32 acquires the first character from the sentence.

  (Step S706) The exclusion unit 32 acquires information indicating whether or not the sentence has a punctuation mark.

  (Step S707) The exclusion unit 32 substitutes 1 for the counter i.

  (Step S708) The exclusion unit 32 determines whether or not the i-th exclusion condition exists in the exclusion condition storage unit 13. If the i-th exclusion condition exists, the process goes to step S709. If the i-th exclusion condition does not exist, the process goes to step S713.

  (Step S709) The exclusion unit 32 acquires the i-th exclusion condition from the exclusion condition storage unit 13.

  (Step S710) The exclusion unit 32 determines whether one or two or more pieces of information acquired in steps S701 to S706 meet the i-th exclusion condition. If the i-th exclusion condition is met, the process goes to step S711. If the i-th exclusion condition is not met, the process goes to step S712.

  (Step S711) The excluding unit 32 substitutes information (for example, “−1”) indicating “exclude” into the variable “exclusion processing result”. Return to upper process.

  (Step S712) The exclusion unit 32 increments the counter i by 1. The process returns to step S708.

  (Step S713) The exclusion unit 32 substitutes information (for example, “0”) indicating “not excluded” into the variable “exclusion process result”. Return to upper process.

  Next, a processing example of the future expression sentence inspection in step S603 will be described using the flowchart of FIG.

  (Step S801) The future expression sentence detection means 331 performs a second filtering process. The second filtering process will be described with reference to the flowchart of FIG.

  (Step S802) The future expression sentence detection means 331 determines whether or not the first result, which is the processing result of step S801, is “undecided”. When the first result is “undecided”, the process goes to step S803, and when the first result is not “undetermined”, the process returns to the upper process.

  (Step S803) The first machine learning means 3311 acquires one or more pieces of information from the sentence to be examined, and constructs a vector. Note that one or more pieces of information for constructing the vector are, for example, one or more of sentence end clue expression information, frequent term information, adverb information, pair information, punctuation information, part of speech information, punctuation information, and sentence attribute information. Information.

  (Step S804) The future expression sentence detection means 331 acquires a prediction result by a machine learning algorithm using the vector configured in step S803 and the first learner. Return to upper process. Here, the prediction result of machine learning is the first result. The first result is information indicating that the sentence is a future expression sentence, or information indicating that the sentence is a non-future expression sentence. The first result may also have a score output by a machine learning function. The score is the likelihood of the first result.

  In the flowchart of FIG. 8, the second filtering process may not exist.

  Next, an example of the second filtering process of step S801 will be described using the flowchart of FIG.

  (Step S901) The future expression sentence detection means 331 substitutes the value “undecided” for the variable “first result”.

  (Step S902) The future expression sentence detection means 331 substitutes 1 for the counter i.

  (Step S903) The future expression sentence detection means 331 determines whether or not the i-th future expression clue expression is stored in the storage unit 1. If the i-th future expression clue expression is stored, the process goes to step S904, and if not stored, the process goes to step S908.

  (Step S904) The future expression sentence detection means 331 acquires the i-th future expression clue expression from the storage unit 1.

  (Step S905) The future expression sentence detection means 331 determines whether or not the i-th future expression clue expression is included in the sentence to be examined. If the i-th future expression clue expression is included, the process goes to step S906, and if not, the process goes to step S907.

  (Step S906) The future expression sentence detection means 331 substitutes information indicating that the value is “future expression sentence” into the variable “first result”. Return to upper process.

  (Step S907) The future expression sentence detection means 331 increments the counter i by 1. The process returns to step S903.

  (Step S908) The future expression sentence detection means 331 substitutes 1 for the counter j.

  (Step S909) The future expression sentence detection means 331 determines whether or not the j-th non-future expression clue expression is stored in the storage unit 1. When the j-th non-future expression clue expression is stored, the process goes to step S910, and when it is not stored, the process returns to the upper process.

  (Step S910) The future expression sentence detection unit 331 acquires the j-th non-future expression clue expression from the storage unit 1.

  (Step S911) The future expression sentence detection means 331 determines whether or not the j-th non-future expression clue expression is included in the sentence to be examined. If the j-th non-future expression clue expression is included, the process goes to step S912. If not, the process goes to step S913.

  (Step S912) The future expression sentence detection means 331 substitutes the value “non-future expression sentence” information for the variable “first result”. Return to upper process.

  (Step S913) The future expression sentence detection means 331 increments the counter j by 1. The process returns to step S909.

  Next, a processing example of the objective means inspection in step S604 will be described using the flowchart of FIG.

  (Step S1001) The second machine learning unit 3321 obtains one or more pieces of information from the sentence to be examined, and constructs a vector. Note that the one or more pieces of information for constructing the vector include, for example, one or more pieces of information out of part-of-speech part-of-speech information, part-of-speech part-of-speech information, punctuation point information, sentence attribute information, and first determination result information.

  (Step S1002) The second machine learning unit 3321 obtains a prediction result by a machine learning algorithm using the vector configured in step S1001 and the second learner. Return to upper process. Here, the prediction result of the machine learning is the second result. The second result is information indicating that the sentence is a purpose means sentence, or information indicating that the sentence is a non-purpose means sentence. The second result may also have a score output by a machine learning function. The score is the likelihood of the second result.

  Next, an example of the acquisition process in step S511 will be described using the flowchart in FIG.

  (Step S1101) The acquisition unit 34 substitutes 1 for a counter i.

  (Step S1102) The acquisition unit 34 determines whether or not the i-th sentence attribute value exists in the sentence attribute values associated with the sentence information. If the i-th sentence attribute value exists, the process goes to step S1103. If the i-th sentence attribute value does not exist, the process goes to step S1107. The i-th sentence attribute value does not overlap with other sentence attribute values.

  (Step S1103) The acquisition unit 34 acquires the number of future-oriented sentences in one or more sentence information corresponding to the i-th sentence attribute value.

  (Step S1104) The acquisition unit 34 calculates a ratio of future-oriented sentences in one or more pieces of sentence information corresponding to the i-th sentence attribute value.

  (Step S1105) The acquisition unit 34 calculates a score using one or more information of the number of future-oriented sentences acquired in step S1103 and the ratio of future-oriented sentences acquired in step S1104.

  (Step S1106) The acquisition unit 34 increments the counter i by 1. The process returns to step S1102.

  (Step S1107) The acquisition unit 34 performs statistical processing using one or more pieces of information among the number of future-oriented sentences acquired in Step S1103, the ratio of future-oriented sentences acquired in Step S1104, and the score acquired in Step S1105. To obtain statistical processing results. Return to upper process. The number of future-oriented sentences acquired in step S1103, the ratio of future-oriented sentences acquired in step S1104, the score acquired in step S1105, and the measurement processing result acquired in step S1107 are examples of future-oriented sentence information.

  Next, an example of the first learning process of step S514 will be described using the flowchart of FIG.

  (Step S1201) The learning unit 31 assigns 1 to the counter i.

  (Step S1202) The learning unit 31 determines whether or not the i-th first learning data exists in the learning data storage unit 12. If the i-th first learning data exists, the process goes to step S1203, and if not, the process goes to step S1206.

  (Step S1203) The learning unit 31 reads the i-th first learning data from the learning data storage unit 12, and constructs a vector using the first learning data.

  (Step S1204) The learning unit 31 acquires a first result paired with the i-th first learning data or a first result included in the i-th first learning data. The first result of the first learning data of the positive example is information indicating that it is a future expression sentence. The first result of the negative first learning data is information indicating that it is not a future expression sentence.

  (Step S1205) The learning unit 31 increments the counter i by 1. The process returns to step S1202.

  (Step S1206) The learning unit 31 uses the vector acquired in step S1203 and the information of the first result set acquired in step S1204 to perform a learning process using a machine learning algorithm to acquire a first learner. .

  (Step S1207) The learning unit 31 accumulates the first learning device acquired in step S1206 in the first learning device storage unit 15. Return to upper process.

  Next, an example of the second learning process in step S516 will be described using the flowchart in FIG.

  (Step S1301) The learning unit 31 substitutes 1 for the counter i.

  (Step S1302) The learning unit 31 determines whether or not the i-th second learning data exists in the learning data storage unit 12. If i-th 2nd learning data exists, it will go to step S1303, and if it does not exist, it will go to step S1306.

  (Step S1303) The learning unit 31 reads the i-th second learning data from the learning data storage unit 12, and constructs a vector using the second learning data.

  (Step S1304) The learning unit 31 acquires a second result paired with the i-th second learning data or a second result included in the i-th second learning data. The second result of the second learning data of the positive example is information indicating that it is a purpose means sentence. The second result of the second learning data of the negative example is information indicating that it is not the objective means sentence.

  (Step S1305) The learning unit 31 increments the counter i by 1. The process returns to step S1302.

  (Step S1306) The learning unit 31 performs a learning process by a machine learning algorithm using the vector acquired in step S1303 and the information of the second result set acquired in step S1304, and acquires a second learner. .

  (Step S1307) The learning unit 31 accumulates the second learning device acquired in step S1306 in the second learning device storage unit 16. Return to upper process.

  Note that another example of the detection process described with reference to FIG. 6 may be an operation as shown in the flowchart of FIG. The flowchart of FIG. 14 includes a part of the steps of the flowchart of FIG.

  Further, another example of the detection process described with reference to FIG. 6 may be an operation as shown in the flowchart of FIG. In the flowchart of FIG. 15, the description of the same steps as those in the flowchart of FIG. 6 is omitted.

  (Step S1501) The detection unit 33 determines whether or not the first result is information indicating “future expression sentence”. If the information indicates “future expression sentence”, the process proceeds to step S604, and if the information does not indicate “future expression sentence”, the process proceeds to step S607.

  (Step S1502) The detection unit 33 determines whether or not the second result is information indicating that “the target means sentence is”. If it is information indicating “is a target means sentence”, the process proceeds to step S606, and if it is not information indicating “is a target means sentence”, the process proceeds to step S607.

  Hereinafter, specific operation examples and experimental results of the information processing apparatus A in the present embodiment will be described. For example, the information processing apparatus A operates according to the procedure shown in FIG. That is, text data 1601 that is text information in the text information storage unit 11 is input, and the exclusion unit 32 performs a first filtering process 1602. Then, the detection unit 33 performs a second filtering process 1603 on a sentence other than the sentence excluded by the exclusion unit 32. Then, the first machine learning means 3311 performs a prediction process by a machine learning algorithm on a sentence that has not been determined whether or not it is a future expression sentence by the second filtering process 1603, and obtains the first result. Obtain (1604). The first result is information indicating whether or not the sentence is a future expression sentence (1605). Further, the second machine learning unit 3321 performs prediction processing on a sentence other than the sentence excluded by the exclusion unit 32 using a machine learning algorithm, and obtains a second result (1606). The second result is information indicating whether or not the sentence is a target means sentence (1607). Then, the selection unit 333 performs matching (1608) and acquires a future-oriented sentence (1609).

  Hereinafter, when the example of text information is a securities report, a specific operation example and experimental results of the information processing apparatus A will be described.

  Assume that the text information storage unit 11 stores, for example, a securities report file shown in FIG.

  Then, it is assumed that the exclusion unit 32 excludes the sentence corresponding to any of the exclusion conditions (1) to (4) described above by the first filtering process (see 1602 in FIG. 16). An example of the removed sentence is shown in FIG.

  In addition, the detection unit 33 uses the second filtering process (see 1603 in FIG. 16) to make a capital investment that includes the sentence “introducing new attractions in a timely manner and taking into account asset efficiency”. "We will implement these measures because we believe it is indispensable." And the sentence "Growth opportunities by developing new products in new markets, such as accelerating the development of immersion-related materials as materials for next-generation semiconductor manufacturing." "We will focus on ensuring that" is a future expression sentence. In addition, the detection unit 33 is developed as a leading comprehensive energy, resource and material group in Asia that has international competitiveness by the management integration through the second filtering process (see 1603 in FIG. 16). We have agreed to aim to contribute to the development of a certain economy and society. "

  Further, the first machine learning means 3311 performs the first prediction process (1604 in FIG. 16) by machine learning using the above-described algorithm using the feature of the number of dimensions shown in FIG. The experimental results in such a case are shown in FIG. In FIG. 19, the accuracy or the like that can be identified as a future expression sentence is a “positive example” line, and the accuracy or the like that can be identified as not a future expression sentence is a “negative example” line. The accuracy of the weighted average of “positive example” and “negative example” is the row of “average (weighted)”. In FIG. 19, on average, the accuracy, the recall, and the F value are all good, that is, less than 0.9. Further, in the negative example, the accuracy was high and the reproduction rate was slightly low, but in the positive example, both exceeded 0.9. Moreover, since the number determined as a negative example was about the same as the number of data of the original negative example, it was not necessarily a high evaluation result because it was determined as a positive example. Further, from FIG. 19, it can be determined that the information processing apparatus A has high determination performance as a whole.

  Further, the second machine learning means 3321 performs the second prediction process (1606 in FIG. 16) by machine learning by the above-described algorithm using the feature of the number of dimensions shown in FIG. The experimental results in such a case are shown in FIG. In FIG. 20, the accuracy or the like that can be specified as the target means sentence is a “positive example” line, and the accuracy or the like that can be specified as not a target means sentence is the “negative example” line. The accuracy of the weighted average of “positive example” and “negative example” is the row of “average (weighted)”. In FIG. 20, on average, the accuracy, recall, and F value were all 0.88, which was a good result. The positive example exceeded 0.9 for all three indices, and the negative example was about 0.75 for all three indices. Since the number determined to be a negative example was about the same as the number of data of the original negative example, it was not necessarily a high evaluation result because it was determined to be a positive example. Further, from FIG. 20, it can be determined that the information processing apparatus A has high determination performance as a whole.

  FIG. 21 shows the experimental results of the accuracy of the information processing apparatus A that extracts the future-oriented sentence by matching the processing in the future expression extraction model and the processing in the objective means extraction model. FIG. 21 shows an evaluation performed using 850 pieces of evaluation data. The evaluation data is data of examples and answers for measuring the performance of the learning device. In FIG. 21, the accuracy or the like that can be identified as a future-oriented sentence is a “positive example” line, and the accuracy or the like that can be identified as not a future-oriented sentence is a “negative example” line. The accuracy of the weighted average of “positive example” and “negative example” is the row of “average (weighted)”. In FIG. 21, on average, accuracy, recall, and F value all exceeded 0.9, which was a very good result. Positive examples exceeded 0.9 for all three indices, and negative examples exceeded 0.8 for all three indices. Since the number determined to be a negative example was about the same as the number of data of the original negative example, it was not necessarily a high evaluation result because it was determined to be a positive example. For negative examples, each evaluation value was significantly improved over the evaluation value of each extracted model by matching the evaluation. Further, as a whole, it can be determined that the information processing apparatus A has very high determination performance.

  The acquisition unit 34 performs statistical processing using the set of future-oriented sentences detected by the detection unit 33, and acquires the future-oriented sentence information shown below. Here, an example of future-oriented sentence information is shown in FIG. In FIG. 22, the future-oriented text information is a statistical processing result using a set of future-oriented texts, and is the ratio of the future-oriented text in the text information to be examined including the securities report of FIG.

  As described above, according to the present embodiment, the future-oriented sentence determination process for determining whether or not the sentence is a future-oriented sentence can be performed with high accuracy.

  Further, according to the present embodiment, it is possible to perform future-oriented sentence determination processing for each sentence included in one or more pieces of sentence information, and output information using the result.

  Further, according to the present embodiment, it is possible to perform future-oriented sentence determination processing at high speed and with high accuracy.

  Furthermore, according to the present embodiment, statistical processing can be performed using the results of future-oriented sentence determination processing.

  Note that the processing in the present embodiment may be realized by software. Then, this software may be distributed by software download or the like. Further, this software may be recorded and distributed on a recording medium such as a CD-ROM. This also applies to other embodiments in this specification. Note that the software that implements the information processing apparatus A in the present embodiment is the following program. That is, this program is a future-oriented sentence that is a future means sentence that is a future expression sentence including that the sentence is related to the future and that has any one or more of a purpose expression and a means expression. A program for functioning as a detection unit that determines whether or not, an acquisition unit that acquires future-oriented sentence information that is information related to a determination result in the detection unit, and an output unit that outputs the future-oriented sentence information acquired by the acquisition unit It is.

  FIG. 23 shows the external appearance of a computer that implements the information processing apparatus A according to various embodiments described above by executing the program described in this specification. The above-described embodiments can be realized by computer hardware and a computer program executed thereon. FIG. 23 is an overview diagram of the computer system 300, and FIG. 24 is a block diagram of the system 300.

  In FIG. 23, a computer system 300 includes a computer 301 including a CD-ROM drive, a keyboard 302, a mouse 303, and a monitor 304.

  24, in addition to the CD-ROM drive 3012, the computer 301 includes an MPU 3013, a bus 3014 connected to the CD-ROM drive 3012, a ROM 3015 for storing a program such as a bootup program, and an MPU 3013. It includes a RAM 3016 for temporarily storing application program instructions and providing a temporary storage space, and a hard disk 3017 for storing application programs, system programs, and data. Although not shown here, the computer 301 may further include a network card that provides connection to a LAN.

  A program that causes the computer system 300 to execute the functions of the information processing apparatus A according to the above-described embodiment may be stored in the CD-ROM 3101, inserted into the CD-ROM drive 3012, and further transferred to the hard disk 3017. Alternatively, the program may be transmitted to the computer 301 via a network (not shown) and stored in the hard disk 3017. The program is loaded into the RAM 3016 at the time of execution. The program may be loaded directly from the CD-ROM 3101 or the network.

  The program does not necessarily include an operating system (OS) or a third-party program that causes the computer 301 to execute the functions of the information processing apparatus A according to the above-described embodiment. The program only needs to include an instruction portion that calls an appropriate function (module) in a controlled manner and obtains a desired result. How the computer system 300 operates is well known and will not be described in detail.

  Further, the computer that executes the program may be singular or plural. That is, centralized processing may be performed, or distributed processing may be performed.

  Further, in each of the above embodiments, it goes without saying that two or more communication means existing in one apparatus may be physically realized by one medium.

  In each of the above embodiments, each process may be realized by centralized processing by a single device, or may be realized by distributed processing by a plurality of devices.

  The present invention is not limited to the above-described embodiments, and various modifications are possible, and it goes without saying that these are also included in the scope of the present invention.

  As described above, the information processing apparatus A according to the present invention is an object in which one or more sentences are future expression sentences including things related to the future, and are sentences having at least one of a purpose expression and a means expression. It has the effect of being able to perform future-oriented sentence determination processing for determining whether or not it is a future-oriented sentence that is a means sentence, and is useful as an information processing apparatus that processes text information such as a securities report.

DESCRIPTION OF SYMBOLS 1 Storage part 2 Reception part 3 Processing part 4 Output part 11 Text information storage part 12 Learning data storage part 13 Exclusion condition storage part 14 End sentence condition storage part 15 1st learning device storage part 16 2nd learning device storage part 31 Learning part 32 Exclusion unit 33 detection unit 34 acquisition unit 331 future expression sentence detection means 332 target means sentence detection means 333 selection means 3311 first machine learning means 3321 second machine learning means

Claims (11)

A detector that determines whether or not the sentence is a future-oriented sentence that is a future expression sentence including a thing related to the future and is a purpose means sentence that is a sentence having any one or more of a purpose expression and a means expression;
An acquisition unit that acquires future-oriented text information that is information about the future-oriented text detected by the detection unit;
An output unit that outputs the future-oriented sentence information acquired by the acquisition unit ;
The detector is
A future expression sentence detection means for determining whether a sentence to be inspected is the future expression sentence;
An objective means sentence detection means for determining whether or not the sentence to be examined is the objective means sentence;
Selecting a future-oriented sentence that is a sentence that the future expression sentence detection means determines to be a future expression sentence and the target means sentence detection means determines to be a purpose means sentence; and
The future expression sentence detection means includes:
Determining whether the sentence is the future expression sentence by performing the first future expression sentence detection process or the second future expression sentence detection process;
The first future expression sentence detection process is:
Whether or not the sentence to be examined is a future expression sentence by using one or more kinds of expressions of the clue expression indicating the future expression stored in the storage unit and the clue expression indicating that it is not the future expression. It is a process to judge,
The second future expression sentence detection process is:
First learning, which is information used to determine whether or not it is a future expression sentence using two or more future expression sentences that are future expression sentences, and is created by a machine learning algorithm Using the first learner stored in the first learner storage unit, from the sentence to be examined, sentence ending clue expression information, frequent term information, adverb information, pair information, punctuation information, part of speech information, Obtain one or more pieces of information of the punctuation information and sentence attribute information, construct a vector having the one or more pieces of information as elements, apply the vector to the first learner, and use the machine learning algorithm to It is a process to determine whether it is an expression sentence,
The first learner is
Vectors obtained using two or more future expression positive example sentences, sentence ending clue expression information relating to a predetermined sentence ending clue expression, frequent term information relating to frequent terms which are predetermined frequent terms, and adverbs included One or more pieces of information: adverb information about a term, pair information about a pair of two clauses at the end of a sentence, part of speech information about a part of speech of a morpheme at the end of a sentence, punctuation information about a punctuation included in a sentence, and sentence attribute information about a sentence attribute Is a learning device obtained by learning with an algorithm of machine learning using two or more vectors.
The objective means sentence detection means comprises:
Determining whether the sentence is the objective means sentence by performing the first objective means sentence detection process or the second objective means sentence detection process;
The first objective means sentence detection process is:
A process of determining whether or not a sentence to be examined is a target means sentence by using a clue expression indicating one or more types of expressions of the purpose expression and means expression stored in the storage unit,
The second purpose means sentence detection process is:
Second learning, which is information used to determine whether a target means sentence is determined by using two or more target means positive example sentences that are target means sentences, and is information created by a machine learning algorithm Using the second learner stored in the second learner storage unit, from the sentence to be inspected, the part-of-speech information before punctuation, the part-of-speech information before the punctuation, the punctuation information, the sentence attribute information, and the first determination result Whether one or more pieces of information are obtained, a vector having the one or more pieces of information as an element is constructed, the vector is applied to the second learner, and the object means sentence is determined by a machine learning algorithm. Is a process of determining whether or not
The second learner is
It is a vector acquired using two or more objective means positive example sentences, part-of-speech part-of-speech information about the part-of-speech of the word before the punctuation point, part-of-speech part-of-speech information about the part-of-speech of the word before the punctuation point, and the punctuation point included in the sentence Two or more vectors whose elements are one or more of the following information: punctuation information, sentence attribute information related to sentence attributes, and first determination result information about whether or not it is a future expression sentence acquired by the first machine learning means It is a learning device that has been acquired by learning using a machine learning algorithm.
The acquisition unit
An information processing apparatus that acquires future-oriented text information that is information related to a future-oriented text selected by the selection means . A sentence information storage unit that stores one or more pieces of sentence information including two or more sentences;
An exclusion condition storage unit that stores an exclusion condition that is a condition for excluding a sentence not corresponding to the future-oriented sentence;
Determining whether or not each of the one or more sentences included in each of the one or more pieces of sentence information satisfies the exclusion condition, and further including an exclusion unit that excludes the sentence that satisfies the exclusion condition;
The detector is
Is one or more sentences in which one or more of each sentence information stored in the sentence information storing portion has one or more statements or et excluding sentences said excluding unit is excluded, in the future expression statement that includes about future The information processing apparatus according to claim 1 , wherein the information processing apparatus detects a future-oriented sentence that is a target means sentence that is a sentence having a target expression or means expression. The exclusion condition is:
One or more of the condition for the number of characters in the sentence, the item in which the sentence appears, the item condition for the item corresponding to the sentence, the attribute condition for the attribute of the sentence, the character condition for the specific character included in the sentence, or The information processing apparatus according to claim 2, which is a combination of two or more conditions. A sentence ending condition storage unit for storing a sentence ending condition that is a condition related to the sentence ending expression;
The first future expression sentence detection process is:
The information according to any one of claims 1 to 3, which is a process for determining that the sentence that satisfies the sentence end condition is the future expression sentence or not the future expression sentence. Processing equipment. The acquisition unit
The detecting unit acquires the number of 1 or more future-oriented statement is detected, performs statistical processing based on the number of the future-oriented statements, statistical processing claims from claim 1 to obtain a future-oriented statement information is the result 4 The information processing apparatus according to any one of claims. One or more pieces of sentence information in the sentence information storage unit in which one or more pieces of sentence information including two or more sentences are stored are associated with sentence attribute values.
The acquisition unit
Obtaining the number of future-oriented sentences for each sentence attribute value, performing statistical processing based on the number of future-oriented sentences, obtaining future-oriented sentence information that is a statistical processing result ,
The future-oriented text information is
6. The information processing apparatus according to claim 5 , wherein the number of future-oriented sentences in the sentence information for each sentence attribute value or the ratio of future-oriented sentences in the sentence information for each sentence attribute value . The sentence attribute value is a company identifier for identifying a company or a business type identifier for identifying a business type,
The acquisition unit
Obtain the number of future-oriented sentences for each company identifier or industry identifier value, perform statistical processing based on the number of future-oriented sentences, obtain the future-oriented sentence information that is the statistical processing result ,
The future-oriented text information is
The information processing apparatus according to claim 6 , wherein the number is the number of future-oriented sentences in the sentence information for each company identifier or industry identifier, or the ratio of future-oriented sentences in the sentence information for each company identifier or industry identifier . The acquisition unit
Using the one or more future-oriented sentences detected by the detection unit, information on the number of future-oriented sentences in the sentence information or the ratio of the future-oriented sentences in the sentence information is acquired, and the future-oriented sentences in the sentence information are acquired. Using the information on the number or the ratio of future-oriented sentences in the sentence information, the future-oriented sentence information that is a score that increases as the number of future-oriented sentences in the sentence information or the ratio of the future-oriented sentences in the sentence information increases. The information processing apparatus according to claim 5 to calculate. The detector is
The information processing apparatus according to any one of claims 1 to 8, wherein the information processing apparatus determines whether or not each of the two or more sentences included in the securities report is a future-oriented sentence. An information processing method realized by a detection unit, an acquisition unit, and an output unit,
The detection unit determines whether or not the sentence is a future expression sentence including a thing related to the future and is a future-oriented sentence that is a target means sentence that is a sentence having any one or more of a purpose expression and a means expression. A detection step;
The acquisition unit, an acquisition step of acquiring a future-oriented statement information, which is information related to the detected contact had been detected future-oriented statements to step,
The output unit comprises an output step of outputting the future-oriented sentence information acquired in the acquisition step ;
The detection step includes
A future expression sentence detection substep for determining whether or not a sentence to be examined is the future expression sentence;
A target means sentence detection substep for determining whether or not a sentence to be examined is the target means sentence;
A selection sub-step for selecting a future-oriented sentence that is a sentence that is determined to be a future expression sentence in the future expression sentence detection sub-step and that is determined to be a target means sentence in the target means sentence detection sub-step; And
In the future expression sentence detection sub-step,
Determining whether the sentence is the future expression sentence by performing the first future expression sentence detection process or the second future expression sentence detection process;
The first future expression sentence detection process is:
Whether or not the sentence to be examined is a future expression sentence by using one or more kinds of expressions of the clue expression indicating the future expression stored in the storage unit and the clue expression indicating that it is not the future expression. It is a process to judge,
The second future expression sentence detection process is:
First learning, which is information used to determine whether or not it is a future expression sentence using two or more future expression sentences that are future expression sentences, and is created by a machine learning algorithm Using the first learner stored in the first learner storage unit, from the sentence to be examined, sentence ending clue expression information, frequent term information, adverb information, pair information, punctuation information, part of speech information, Obtain one or more pieces of information of the punctuation information and sentence attribute information, construct a vector having the one or more pieces of information as elements, apply the vector to the first learner, and use the machine learning algorithm to It is a process to determine whether it is an expression sentence,
The first learner is
Vectors obtained using two or more future expression positive example sentences, sentence ending clue expression information relating to a predetermined sentence ending clue expression, frequent term information relating to frequent terms which are predetermined frequent terms, and adverbs included One or more pieces of information: adverb information about a term, pair information about a pair of two clauses at the end of a sentence, part of speech information about a part of speech of a morpheme at the end of a sentence, punctuation information about a punctuation included in a sentence, and sentence attribute information about a sentence attribute Is a learning device obtained by learning with an algorithm of machine learning using two or more vectors.
In the objective means sentence detection sub-step,
Determining whether the sentence is the objective means sentence by performing the first objective means sentence detection process or the second objective means sentence detection process;
The first objective means sentence detection process is:
A process of determining whether or not a sentence to be examined is a target means sentence by using a clue expression indicating one or more types of expressions of the purpose expression and means expression stored in the storage unit,
The second purpose means sentence detection process is:
Second learning, which is information used to determine whether a target means sentence is determined by using two or more target means positive example sentences that are target means sentences, and is information created by a machine learning algorithm Using the second learner stored in the second learner storage unit, from the sentence to be inspected, the part-of-speech information before punctuation, the part-of-speech information before the punctuation, the punctuation information, the sentence attribute information, and the first determination result Whether one or more pieces of information are obtained, a vector having the one or more pieces of information as an element is configured, the vector is applied to the second learning device, and the object means sentence is determined by a machine learning algorithm. Is a process of determining whether or not
The second learner is
It is a vector acquired using two or more objective means positive example sentences, and the part-of-speech part-of-speech information about the part-of-speech of the word before the punctuation point, the part-of-speech part-of-speech information about the part-of-speech of the word before the punctuation point, and the punctuation point included in the sentence Two or more vectors whose elements are one or more of the following information: punctuation information, sentence attribute information related to sentence attributes, and first determination result information about whether or not it is a future expression sentence acquired by the first machine learning means It is a learning device that has been acquired by learning using a machine learning algorithm.
In the obtaining step,
An information processing method for acquiring future-oriented sentence information that is information related to a future-oriented sentence selected in the selection sub-step . Computer
A detector that determines whether or not the sentence is a future-oriented sentence that is a future expression sentence including a thing related to the future and is a purpose means sentence that is a sentence having any one or more of a purpose expression and a means expression;
An acquisition unit that acquires future-oriented text information that is information about the future-oriented text detected by the detection unit;
A program for functioning as an output unit that outputs the future-oriented sentence information acquired by the acquisition unit ,
The detector is
A future expression sentence detection means for determining whether a sentence to be inspected is the future expression sentence;
An objective means sentence detection means for determining whether or not the sentence to be examined is the objective means sentence;
A selection means for selecting a future-oriented sentence that is a sentence that the future expression sentence detection means determines to be a future expression sentence and the target means sentence detection means determines to be a target means sentence; Make the computer function,
The future expression sentence detection means includes:
Determining whether the sentence is the future expression sentence by performing the first future expression sentence detection process or the second future expression sentence detection process;
The first future expression sentence detection process is:
A process of determining whether or not the sentence to be examined is a future expression sentence using a clue expression indicating the future expression.
The second future expression sentence detection process is:
First learning, which is information used to determine whether or not it is a future expression sentence using two or more future expression sentences that are future expression sentences, and is created by a machine learning algorithm Using the first learner stored in the first learner storage unit, from the sentence to be examined, sentence ending clue expression information, frequent term information, adverb information, pair information, punctuation information, part of speech information, Obtain one or more pieces of information of the punctuation information and sentence attribute information, construct a vector having the one or more pieces of information as elements, apply the vector to the first learner, and use the machine learning algorithm to It is a process to determine whether it is an expression sentence,
The first learner is
Vectors obtained using two or more future expression positive example sentences, sentence ending clue expression information relating to a predetermined sentence ending clue expression, frequent term information relating to frequent terms which are predetermined frequent terms, and adverbs included One or more pieces of information: adverb information about a term, pair information about a pair of two clauses at the end of a sentence, part of speech information about a part of speech of a morpheme at the end of a sentence, punctuation information about a punctuation included in a sentence, and sentence attribute information about a sentence attribute Is a learning device obtained by learning with an algorithm of machine learning using two or more vectors.
The objective means sentence detection means comprises:
Determining whether the sentence is the objective means sentence by performing the first objective means sentence detection process or the second objective means sentence detection process;
The first objective means sentence detection process is:
A process of determining whether or not a sentence to be examined is a purpose means sentence using a clue expression indicating one or more types of expressions among the purpose expression and means expression,
The second purpose means sentence detection process is:
Second learning, which is information used to determine whether a target means sentence is determined by using two or more target means positive example sentences that are target means sentences, and is information created by a machine learning algorithm Using the second learner stored in the second learner storage unit, from the sentence to be inspected, the part-of-speech information before punctuation, the part-of-speech information before the punctuation, the punctuation information, the sentence attribute information, and the first determination result Whether one or more pieces of information are obtained, a vector having the one or more pieces of information as an element is configured, the vector is applied to the second learning device, and the object means sentence is determined by a machine learning algorithm. Is a process of determining whether or not
The second learner is
It is a vector acquired using two or more objective means positive example sentences, and the part-of-speech part-of-speech information about the part-of-speech of the word before the punctuation point, the part-of-speech part-of-speech information about the part-of-speech of the word before the punctuation point, and the punctuation point included in the sentence Two or more vectors whose elements are one or more of the following information: punctuation information, sentence attribute information related to sentence attributes, and first determination result information about whether or not it is a future expression sentence acquired by the first machine learning means It is a learning device that has been acquired by learning using a machine learning algorithm.
The acquisition unit
A program for causing the computer to function as acquiring future-oriented text information that is information related to a future-oriented text selected by the selection means .