JP2019109615A - Classification device, learning device, classification method, learning method, and computer program - Google Patents

Abstract

To provide a classification device, a learning device, a classification method, a learning method, and a computer program capable of further easily classifying information.SOLUTION: A classification device includes: a morpheme analysis unit which divides a character string into morphemes; and a first classification information determination unit which performs prescribed machine learning on the basis of a plurality of pieces of training data including the character strings and classification results of the character strings so as to obtain first learning result data from a learning result data storage unit which records the first learning result data including frequency by which the first classification information representing the feature of the character string has been given to the character string, and determines the first classification information of the character strings on the basis of the first learning result data and the morphemes.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a classification device, a learning device, a classification method, a learning method, and a computer program.

  The receipt issued by buying and selling the item describes the item name and the sales amount. By acquiring information on products such as product names and sales amounts from the collected receipts and classifying them, it is possible to grasp the sales and sales of products of competitors. Therefore, a master has been constructed and maintained for storing the result of classifying information on goods.

JP, 2011-103038, A

  However, in the above method, since it is necessary to update the master each time a new product is put on the market, construction and maintenance of the master may take a lot of effort. In addition, even if the same product is classified as a different product, it may be erroneously classified as a different product due to writing fluctuation such as katakana writing or hiragana writing for each receipt. In such a case, there has been a case where the information on the product can not be correctly classified unless the master is updated or the classification error of the product is corrected individually.

  In view of the above circumstances, the present invention has an object to provide a classification device, a learning device, a classification method, a learning method, and a computer program that can classify information more easily.

  In one aspect of the present invention, the character string is obtained by performing predetermined machine learning based on a plurality of training data including a morphological analysis unit that divides a character string into morphemes, and a character string and a classification result of the character string. The first learning result data is acquired from the learning result data storage unit that records the first learning result data including the frequency at which the first classification information representing the feature of the character string has been added to the character string so far, and the first learning result And a first classification information determination unit that determines the first classification information of the character string based on data and the morpheme.

  One embodiment of the present invention is the classification device described above, further comprising a morpheme pair determination unit that determines a morpheme pair including any two of the morphemes based on a predetermined condition, and the first classification information The determination unit performs predetermined machine learning based on the training data, and the number of times the first classification information has been given to the character string based on morphemes included in morpheme pairs of the training data. Acquiring the second learning result data from the learning result data storage unit that records the second learning result data including the second learning result data, and based on the second learning result data and the morpheme pair, the first classification information of the character string Decide.

  One embodiment of the present invention is the classification device described above, wherein, by performing predetermined machine learning based on the training data, a plurality of second classification information items associated with the determined first classification information. Acquiring third learning result data from the learning result data storage unit that records third learning result data for determining second classification information representing characteristics of the character string, and the third learning result data and the third learning result data The information processing apparatus further includes a second classification information determination unit that determines the second classification information based on morphemes.

  One embodiment of the present invention is the classification device described above, further comprising a morpheme substitution unit that substitutes the morpheme with another morpheme based on a predetermined condition.

  One embodiment of the present invention is the classification device described above, wherein a morpheme having a meaning closest to the character string is determined among the morphemes, and a morpheme pair including the specified morpheme and another morpheme having a meaning similar is determined. The first classification information determination unit further performs predetermined machine learning based on the training data, whereby the first classification information determination unit performs the first classification information determination based on the morpheme included in the morpheme pair of the training data. The second learning result data is acquired from the learning result data storage unit that records the second learning result data including the number of times classification information has been added to the character string so far, the second learning result data and the morpheme The first classification information of the character string is determined based on the pair.

  One aspect of the present invention acquires a plurality of training data including a morphological analysis unit that divides a character string into morphemes, a character string and a classification result of the character string, and performs predetermined machine learning to obtain the character. And a learning result data generation unit that generates first learning result data including the frequency at which the first classification information representing the feature of the string has been assigned to the character string so far.

  In one aspect of the present invention, a classification device divides a character string into morphemes, a morphological analysis step; and a classification device includes predetermined character learning based on a plurality of training data including a character string and a classification result of the character string. The first learning result data is output from the learning result data storage unit that records the first learning result data including the frequency at which the first classification information representing the feature of the character string has been added to the character string by doing A first classification information determination step of acquiring the first classification information of the character string based on the first learning result data and the morpheme.

  In one aspect of the present invention, the learning device divides a character string into morphemes, and the learning device acquires a plurality of training data including a character string and a classification result of the character string, and a predetermined machine A learning result data generation step of generating first learning result data including a frequency at which the first classification information representing the feature of the character string has been added to the character string by performing learning It is.

  One embodiment of the present invention is a computer program for causing a computer to function as the classification device described above.

  One embodiment of the present invention is a computer program for causing a computer to function as the above-described learning device.

  The present invention makes it possible to more easily classify information.

It is a functional block diagram showing the functional composition of information processor 100 of a 1st embodiment. It is a figure which shows one specific example of a training data table. It is a figure which shows one specific example of a substitution data table. It is a figure which shows one specific example of a classification | category information table. It is a figure which shows one specific example of a master information table. It is a figure which shows one specific example of morphological analysis. It is a figure which shows one specific example of the 1st category matrix produced | generated based on the morpheme. It is a figure which shows one specific example of the 1st category matrix produced | generated based on the morpheme pair. It is a figure which shows one specific example of the 2nd category matrix produced | generated based on the morpheme. It is a figure which shows one specific example of the 2nd category matrix produced | generated based on the morpheme pair. It is a flow chart which shows a flow of processing of generation of a category matrix of information processor 100 of a 1st embodiment. It is a flow chart which shows a flow of classification processing of information processor 100 of a 1st embodiment. It is a flow chart which shows a flow of processing of determination of classification information of information processor 100 of a 1st embodiment. It is a flow chart which shows a flow of processing of determination of classification information of information processor 100 of a 1st embodiment. It is a functional block diagram showing the functional composition of information processor 100a of a 2nd embodiment.

  FIG. 1 is a functional block diagram showing a functional configuration of the information processing apparatus 100 according to the first embodiment. The information processing apparatus 100 is an apparatus such as a personal computer, a server, a smartphone or a tablet computer. The information processing apparatus 100 adds information indicating characteristics of arbitrary character string information to the input character string information. The character string information may be any information as long as it is a text, a word or a character such as a number. The character string information of the present embodiment will be described, for example, on the assumption that a receipt issued by trading is read by an optical character recognition (OCR). In the present embodiment, the character string information is described as information including at least a product name, sales number, and sales amount related to trading, but is not limited thereto. The character string information may be, for example, a character string input by the user, or may be a character string acquired from the Internet or the like.

  The information processing apparatus 100 includes a processor, a memory, an auxiliary storage device, and the like connected by a bus, and executes the classification program or the learning program to thereby execute the communication unit 101, the input unit 102, the display unit 103, and the training data storage unit 104. Functions as an apparatus including the replacement data storage unit 105, the classification information storage unit 106, the category matrix storage unit 107, the master information storage unit 108, and the control unit 109. Note that all or part of the functions of the information processing apparatus 100 may be realized using hardware such as an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA). . The classification program or the learning program program may be recorded on a computer readable recording medium. The computer readable recording medium is, for example, a portable medium such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, or a storage device such as a hard disk built in a computer system. The classification program or the learning program program may be transmitted via a telecommunication line.

  The communication unit 101 is a network interface. The communication unit 101 communicates with an external device via a network. The external apparatus may be, for example, an apparatus such as a personal computer, a server, a smartphone, a tablet computer, or the like, or may be an image forming apparatus, an image processing apparatus, or the like. The communication unit 101 may communicate using a communication method such as wireless LAN (Local Area Network), wired LAN, Bluetooth (registered trademark), or Long Term Evolution (LTE) (registered trademark), for example.

  The input unit 102 is configured using an input device such as a touch panel, a mouse, and a keyboard. The input unit 102 may be an interface for connecting an input device to the information processing apparatus 100. In this case, the input unit 102 generates input data (for example, instruction information indicating an instruction to the information processing apparatus 100) from an input signal input to the input device, and inputs the input data to the information processing apparatus 100.

  The display unit 103 is an output device such as a cathode ray tube (CRT) display, a liquid crystal display, and an organic electro luminescence (EL) display. The display unit 103 may be an interface for connecting an output device to the information processing apparatus 100. In this case, the display unit 103 generates a video signal from the video data and outputs the video signal to a video output device connected to itself.

  The training data storage unit 104 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The training data storage unit 104 stores a training data table. The training data table is a table for recording character string information used to generate learning result data. It is a table which records the 1st classification information and the 2nd classification information given to the inputted character string information.

  FIG. 2 is a diagram showing one specific example of the training data table. The training data table has training data records. The training data record has values of year, month, store name, product name, first classification information, second classification information, sales amount and sales number. Year and month are information included in the string information. The year and month represent the year and month when the sale was made. The store name is information included in the character string information. The store name represents the name of the store that has been bought and sold. The product name is information included in the character string information. The item name represents the name of the item for sale. The product name may include characters that are misrecognized by the OCR. The first classification information indicates in which category the product name is a product. The second classification information represents information that more specifically describes the product name based on the first classification information. The first classification information and the second classification information are associated in a 1: n relationship. The first classification information and the second classification information may be designated by the user via the input unit 102. The combination of the first classification information and the second classification information is uniquely determined. Sales represents the sold amount of the product represented by the product name. The number of sales represents the number of products sold in the product name.

  In the example shown in FIG. 2, in the training data record at the top of the training data table, the year-month value is "201706", the store name value is "store 200", and the product name value is "d". The value of the first classification information is "food", the value of the second classification information is "rice", the value of sales is "330", and the value of the number of sales is "3". Therefore, according to the training data record at the top of the training data table, the year and month of the input character string information is "June 2017", the store name is "store 200", and the product name is "d" In the case of rice ball rice ball Hidaka kelp, it is understood that "food" is given as the first classification information and "rice" is given as the second classification information. In addition, it can be seen that sales of the trade name "d Teshima rice ball, Hidaka kelp" is "330" and the number of sales is "3". The training data table shown in FIG. 2 is only one specific example. Therefore, the training data table may be configured in a mode different from that of FIG. For example, the training data table may not have the second classification information or may be configured to have the third classification information.

  Returning to FIG. 2, the description of the information processing apparatus 100 will be continued. The replacement data storage unit 105 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The replacement data storage unit 105 stores a replacement data table. The replacement data table is a table that records rules for replacing characters with different characters.

  FIG. 3 is a diagram showing a specific example of the replacement data table. The replacement data table has replacement data records. The replacement data record has each value before and after replacement. Before substitution represents the character before substitution. After substitution, the character before substitution represents the character to be substituted.

  In the example shown in FIG. 3, in the replacement data record at the top of the replacement data table, the value before replacement is “manual” and the value after replacement is “handroll”. Therefore, according to the replacement data record at the top of the replacement data table, it can be understood that the "manual" is replaced with the "handroll" when the given character is the "manual". The replacement data table shown in FIG. 3 is only one specific example. Therefore, the replacement data table may be configured in a mode different from that of FIG.

  Returning to FIG. 2, the description of the information processing apparatus 100 will be continued. The classification information storage unit 106 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The classification information storage unit 106 stores a classification information table. The classification information table is a table for recording the correspondence between the first classification information and the second classification information.

  FIG. 4 is a diagram showing a specific example of the classification information table. The classification information table has classification information records. The classification information record has values of the first classification information and the second classification information. In the example shown in FIG. 4, in the separation information record at the top of the classification information table, the value of the first classification information is “food”, the value of the second classification information is “rice”, “noodles”, “bread” "," Meats "," Vegetables ", .... Therefore, according to the classification information record at the top of the classification information table, when "food" is given as the first classification information, "rice", "noodles", "breads", It can be seen that any of "meats", "vegetables", ... is applied. The classification information table shown in FIG. 4 is only one specific example. Therefore, the classification information table may be configured in a mode different from FIG.

  Returning to FIG. 2, the description of the information processing apparatus 100 will be continued. The category matrix storage unit 107 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The category matrix storage unit 107 stores a first category matrix or a second category matrix. The first category matrix is a matrix that holds values associated with the first determination information for determining the first classification information given to the character string information. The first determination information is information representing the frequency of the classification result of the first classification information added to the character string information included in the training data. The second category matrix is a matrix that holds values associated with second determination information for determining second classification information to be attached to character string information. The second determination information is information representing the frequency of the classification result of the second classification information added to the character string information included in the training data. The detailed description of the first category matrix and the second category matrix will be described later. The category matrix storage unit 107 is an aspect of the learning result data storage unit.

  The master information storage unit 108 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The master information storage unit 108 stores a master information table. The master information table is a table in which the correspondence between the character string information and the first classification information or the second classification information attached to the character string information is recorded.

  FIG. 5 is a diagram showing a specific example of the master information table. The master information table has master information records. A master information record has each value of goods name, 1st classification information, and 2nd classification information. The product name represents the name of the product. The product name is an aspect of character string information. The first classification information represents first classification information attached to the item name. The second classification information represents second classification information attached to the product name.

  In the example shown in FIG. 5, in the master information record at the top of the master information table, the value of the product name is "handroll rice ball Hidaka kelp", the value of the first classification information is "food", and the second classification information The value is "rice". Therefore, according to the master information record at the top of the master information table, when the given character string information is "handroll rice ball Hidaka kelp", the first classification information is "food" and the second classification information is It turns out that it is "rice". The master information table shown in FIG. 5 is only one specific example. Therefore, the master information table may be configured in a mode different from FIG. For example, the master information table may be configured to store the number of times the product name is given as character string information.

  Returning to FIG. 2, the description of the information processing apparatus 100 will be continued. The control unit 109 controls the operation of each unit of the information processing apparatus 100. The control unit 109 is executed by an apparatus including a processor and a memory, for example. The control unit 109 functions as a data acquisition unit 110, a morpheme analysis unit 111, a morpheme replacement unit 112, a morpheme pair determination unit 113, a learning unit 114, and a classification unit 115 by executing a classification program or a learning program.

  The data acquisition unit 110 acquires character string information. The data acquisition unit 110 may receive character string information via the input unit 102 or may receive it from the communication unit 101. For example, the data acquisition unit 110 may acquire, as character string information, information obtained by reading a sales receipt or the like by OCR. The data acquisition unit 110 outputs the acquired character string information to the classification unit 115. The data acquisition unit 110 may record the acquired character string information in the training data storage unit 104.

  The morphological analysis unit 111 performs morphological analysis on character string information. Morphological analysis is processing to decompose character string information into morphemes. Morphological analysis is a process of determining part of speech with respect to decomposed morphemes. Known algorithms may be used for morphological analysis. The morpheme is obtained by dividing the string information. A morpheme is a character of the smallest unit with some meaning.

  The morpheme replacing unit 112 replaces a predetermined morpheme with a different morpheme. The predetermined morpheme is a morpheme held in the pre-replacement column of the replacement data table. The morpheme replacing unit 112 determines whether the morpheme is stored in the pre-replacement column of the replacement data table. When the morpheme is stored in the pre-replacement column, the morpheme replacement unit 112 replaces the morpheme with the morpheme of the post-replacement column. If the morpheme is not stored in the pre-replacement column, the morpheme replacement unit 112 does not replace the morpheme with a different morpheme.

  FIG. 6 is a diagram showing a specific example of morphological analysis. In FIG. 6, the case where "d hand-made rice ball Nidaka kelp" is acquired as character string information is demonstrated. The morphological analysis unit 111 divides character string information into morphemes. The morphological analysis unit 111 decomposes “d hand-roll rice ball Hidaka kelp” into “d”, “hand-rice ball”, “rice ball”, “hidaka” and “kelp” morphemes. The morpheme replacing unit 112 determines whether the morpheme is stored in the pre-replacement column of the replacement data table. According to FIGS. 3 and 6, the morpheme substitution unit 112 substitutes the morpheme "hand-roll" with "hand-roll". The morphological analysis unit 111 determines the part of speech for each morpheme. According to FIG. 6, the morphological analysis unit 111 determines that d is “symbol”, handroll is “noun”, rice ball is “noun”, Hidaka is “noun”, and kelp is “noun”. Determine. The morpheme analysis unit 111 deletes the morpheme determined as the symbol.

  Returning to FIG. 2, the description of the information processing apparatus 100 will be continued. The morpheme pair determining unit 113 determines morpheme pairs. The morpheme pair is a combination of morphemes positioned at the beginning and end of the character string information among the decomposed morphemes. For example, when the character string information is "d hand-rolled rice ball Hidaka kelp", the morphemes are "d", "hand-roll", "rice ball", "hidaka" and "kelp". “D” is deleted because it is a symbol. Therefore, the morpheme pair determining unit 113 determines “hand-roll” and “kobu” as morpheme pairs. The morpheme pair determining unit 113 determines a morpheme pair based on the morpheme received from the morpheme analyzing unit 111.

  The learning unit 114 functions as a first category matrix generation unit 116 and a second category matrix generation unit 117 by executing a learning program. The classification unit 115 functions as a master information search unit 118, a first classification information determination unit 119, a second classification information determination unit 120, and a master information generation unit 121 by executing a classification program.

  The first category matrix generation unit 116 generates a first category matrix by predetermined machine learning based on a morpheme or a morpheme pair. As the morpheme, a morpheme decomposed by the morpheme analysis unit 111 may be used. The first category matrix is a matrix that represents the frequency of the first classification information attached to the morpheme. The first category matrix generated by the present machine learning is an aspect of the first learning result data. The morpheme pair determined by the morpheme pair determining unit 113 may be used as the morpheme pair. First, machine learning in the case where the first category matrix generation unit 116 generates a first category matrix based on morphemes will be described.

  The first category matrix generation unit 116 acquires first classification information from the classification information table stored in the classification information storage unit 106. The first category matrix generation unit 116 acquires morphemes from the morpheme analysis unit 111. The morpheme analysis unit 111 generates a morpheme by performing a morpheme analysis on the product name column of the training data table stored in the training data storage unit 104. The morphological analysis unit 111 outputs the generated morpheme to the first category matrix generation unit 116. The first category matrix generation unit 116 sets the first classification information as a row of the first category matrix and the morpheme as a column of the first category matrix.

  The first category matrix generation unit 116 counts the number of acquired morphemes. For example, when the number of acquired morpheme “handrolls” from the morphological analysis unit 111 is “3”, the first category matrix generation unit 116 counts the number of morphemes “handrolls” as “3”. The first category matrix generation unit 116 calculates, for each morpheme, the proportion of morphemes belonging to each piece of first classification information. As the value of the first classification information, the first classification information associated with the product name is used in the training data table. Specifically, for each morpheme, first category matrix generation section 116 calculates the ratio by dividing the number of morphemes associated with each piece of first classification information by the total number of the morphemes counted. . For example, the total number of morpheme "handroll" is 7, among which morpheme "handroll" associated with the first classification information "food" is 6, morpheme associated with the first classification information "book" When the “hand-roll” is 1, the first category matrix generation unit 116 determines that the proportion of morphemes belonging to “food” is “0.86”, the proportion of morphemes belonging to “books” is “0.14”, The proportion of morphemes belonging to the other first classification information is calculated as “0.00”. The first category matrix generation unit 116 sets the calculated value in each cell of the first category matrix. The first category matrix generation unit 116 records the first category matrix in the category matrix storage unit 107. The first category matrix generation unit 116 may be configured to consider the number of sales when calculating the ratio. For example, when “3” is recorded as the number of sales in the training data table, the first category matrix generation unit 116 may count the number of morphemes as “3”.

  FIG. 7 is a diagram showing a specific example of the first category matrix generated based on morphemes. Rows of the first category matrix represent first classification information such as “morpheme”, “food”, “beverage”, “book”, “stationery”, “daily goods”, and so on. The columns of the first category matrix represent morphemes such as "salt", "literary arts", "cutlet", "banana", "ramen", and so on.

  In the example shown in FIG. 7, in the top row of the first category matrix, the morpheme value is “salt”, the food value is “0.42”, the beverage value is “0.33”, and the book value Is “0.01”, the value of stationery is “0.02”, the value of daily necessities is “0.04”,. Therefore, according to the top row of the first category matrix, when the morpheme contains "salt", the possibility of being a food is "0.42", the possibility of being a beverage is "0.33", and One possibility is "0.01", the possibility of being a stationery is "0.02", the possibility of being a commodity is "0.04", and so on.

  Next, machine learning in the case where the first category matrix generation unit 116 generates a first category matrix based on morpheme pairs will be described. The first category matrix generated by the present machine learning is an aspect of the second learning result data. The first category matrix generation unit 116 acquires morpheme pairs from the morpheme pair determination unit 113. The first category matrix generation unit 116 sets the first classification information as the rows of the first category matrix, and sets the morpheme pairs as the columns of the first category matrix.

  The first category matrix generation unit 116 counts the number of acquired morpheme pairs. For example, when the first category matrix generation unit 116 determines that the number of morpheme pairs “hand-rolled” and “Konbu” acquired is “3”, the number of morpheme pairs “hand-rolled” and “kelp” is “3”. Count The first category matrix generation unit 116 calculates, for each morpheme pair, a ratio of morpheme pairs belonging to each piece of first classification information. Specifically, for each morpheme pair, first category matrix generation section 116 divides the number of morpheme pairs associated with each piece of first classification information by the total number of counted morpheme pairs. , Calculate the ratio. For example, when the total number of morpheme pairs "hand-roll" and "kelp" is 3, and the morpheme pair "hand-roll" and "kelp" corresponding to the first classification information "food" are 3, for example, " The proportion of morpheme pairs belonging to "food" is "1.00", and the proportion of morpheme pairs belonging to the other first classification information is "0.00". The first category matrix generation unit 116 sets the calculated ratio to each cell of the first category matrix. The first category matrix generation unit 116 records the first category matrix in the category matrix storage unit 107. The first category matrix generation unit 116 may be configured to consider the number of sales when calculating the ratio. For example, when “3” is recorded as the number of sales in the training data table, the first category matrix generation unit 116 may count the number of morphemes as “3”.

  FIG. 8 is a diagram showing a specific example of the first category matrix generated based on morpheme pairs. The rows of the first category matrix indicate the first classification information such as “first morpheme”, “second morpheme”, “food”, “drink”, “book”, “stationery”, “daily necessities”,. expressed. The columns of the first category matrix are the columns of the first morpheme "company A", "company A", "company A", "company A", "company B", etc., and the columns of the second morpheme "noodle", " Yakisoba "," mini "," pen ", etc. are represented. The first morpheme represents the top morpheme of the morpheme pair. The second morpheme represents the morpheme at the end of the morpheme pair.

  In the example shown in FIG. 8, in the top row of the first category matrix, the value of the first morpheme is “A company”, the value of the second morpheme is “noodle”, the value of the food is “0.97”, The beverage value is "0.01", the book value is "0.01", the stationery value is "0.00", the daily necessities value is "0.00", and so on. Therefore, according to the top row of the first category matrix, when the morpheme pair is "company A" and "noodle", the possibility of being a food is "0.97" and the possibility of being a beverage is "0. 01 ", the possibility of being a book is" 0.01 ", the possibility of being a stationery is" 0.00 ", the possibility of being a daily necessities is" 0.00 ", and so on.

  The second category matrix generation unit 117 generates a second category matrix by predetermined machine learning based on the morpheme or the morpheme pair and the first category matrix generated by the first category matrix generation unit 116. As the morpheme, a morpheme decomposed by the morpheme analysis unit 111 may be used. The second category matrix is a matrix that represents the frequency of the second classification information assigned to the morpheme. The second category matrix generated by the present machine learning is an aspect of the third learning result data. The morpheme pair determined by the morpheme pair determining unit 113 may be used as the morpheme pair. First, machine learning in the case where the second category matrix generation unit 117 generates a second category matrix based on morphemes will be described.

  The second category matrix generation unit 117 acquires, from the classification information table stored in the classification information storage unit 106, the second classification information associated with the determined first classification information. The second category matrix generation unit 117 acquires morphemes from the morpheme analysis unit 111. The second category matrix generation unit 117 sets second classification information as a row of the first category matrix and a morpheme as a column of the second category matrix.

  The second category matrix generation unit 117 counts the number of acquired morphemes for each second classification information. For example, if the number of acquisitions of the morpheme "handroll" whose second classification information is "rice" is "1", the second category matrix generation unit 117 generates second classification information "of the morpheme" handroll ". The number of “rices” is counted as “1”. Furthermore, if the number of acquisitions of the morpheme “handroll” whose second classification information is “noodles” is “0”, the second category matrix generation unit 117 generates the second classification information “noodles” of the morpheme “handroll”. The number of "" counts as "0". The second category matrix generation unit 117 sets the counted value to each cell of the second category matrix. The second category matrix generation unit 117 records the second category matrix in the category matrix storage unit 107. The second category matrix generation unit 117 may be configured to consider the number of sales when counting. For example, if “3” is recorded as the number of sales of morpheme “handroll” whose second classification information is “rice” in the training data table, second category matrix generation section 117 sets the number of morphemes to “ It may be counted as 3 ".

  FIG. 9 is a diagram showing a specific example of the second category matrix generated based on morphemes. Rows of the second category matrix correspond to the first classification information “food” such as “morpheme”, “rice”, “noodles”, “breads”, “meats”, “vegetables”,. The second classified information is represented. The columns of the first category matrix represent morphemes such as “yakisoba”, “chishu”, “tonkotsu”, “28”, “meet”, and so on.

  In the example shown in FIG. 9, in the top row of the first category matrix, the morpheme value is "Yakisoba", the rice value is "0", the noodle value is "12", and the bread value is " 6 ", the value of meat is" 2 ", the value of vegetables is" 3 ", and so on. Therefore, according to the top row of the first category matrix, if the morpheme contains "fried soba", the possibility of being rice is "0", the possibility of being noodles is "12", and it is possible to be bread The sex is "6", the possibility of being meat is "2", the possibility of being vegetables is "3", and so on.

  Next, machine learning in the case where the second category matrix generation unit 117 generates a second category matrix based on morpheme pairs will be described. The second category matrix generated by the present machine learning is an aspect of the fourth learning result data. The second category matrix generation unit 117 acquires morpheme pairs from the morpheme pair determination unit 113. The second category matrix generation unit 117 sets second classification information as a row of the second category matrix and sets a morpheme pair as a column of the second category matrix.

  The second category matrix generation unit 117 counts the number of acquired morpheme pairs for each second classification information. The second classification information is classification information associated with the first classification information attached to the character string information related to the acquired morpheme pair. For example, if the second category information is “rices”, the morpheme pair “handroll”, and the number of acquired “Kunbu” is “1”, the second category matrix generation unit 117 determines that the second category information is “rice”. The number of morpheme pairs "handroll" and "kelp" related to "class" is counted as "1". In addition, when the second category information generation unit 117 acquires the second classification information “noodles”, the morpheme pair “handroll”, and “kelp” but is “0”, the second classification information relates to “noodles”. The number of morpheme pairs "handroll" and "kelp" is counted as "0". The second category matrix generation unit 117 sets the counted value as a second category matrix. The second category matrix generation unit 117 records the second category matrix in the category matrix storage unit 107. The second category matrix generation unit 117 may be configured to consider the number of sales when counting. For example, if “3” is recorded as the number of sales of morpheme “handroll” whose second classification information is “rice” in the training data table, second category matrix generation section 117 sets the number of morphemes to “ It may be counted as 3 ".

  FIG. 10 is a diagram showing a specific example of the second category matrix generated based on morpheme pairs. The rows of the second category matrix are the first morphemes, the second morphemes, the rices, the noodles, the breads, the meats, the vegetables, etc. Second classification information associated with the classification information is represented. The columns of the first category matrix have columns of the first morpheme "company A", "company A", "company A", "company A", "company A", "company B", etc. "Noodles", "Yosoba", "Mini", "Soy sauce" ... etc. are displayed. The first morpheme represents the top morpheme of the morpheme pair. The second morpheme represents the morpheme at the end of the morpheme pair.

  In the example shown in FIG. 9, in the top row of the first category matrix, the value of the first morpheme is “A company”, the value of the second morpheme is “noodle”, the value of rice is “0”, and noodles The value of “9”, the value of bread is “0”, the value of meat is “0”, the value of vegetables is “0”,. Therefore, according to the top row of the second category matrix, if the morpheme pair is "company A" or "noodle", the possibility of being rice is "0", the possibility of being noodles is "9", It is understood that the possibility of being bread is “0”, the possibility of being meat is “0”, the possibility of being vegetables is “0”,.

  The master information search unit 118 determines whether character string information is included in a predetermined column included in the master information table. When the character string recorded in the predetermined column matches the character string information, the master information search unit 118 determines that the character string information is included. When the character string recorded in the predetermined column and the character string information do not match, the master information search unit 118 determines that the character string information is not included. The predetermined columns are previously designated by the user. The user may be, for example, a purchaser of a product or a person who operates the information processing apparatus 100. In the case of the present embodiment, the predetermined column is the "product name" column provided in the training data table. The number of predetermined columns may be one or more.

  The first classification information determination unit 119 determines first classification information for the character string information based on the morphemes or morpheme pairs and the first category matrix stored in the category matrix storage unit 107. First, the case where the first classification information determination unit 119 determines the first classification information based on the first category matrix based on the morpheme will be described. The first classification information determination unit 119 specifies a row of morphemes analyzed by the morpheme analysis unit 111 from the first category matrix. The first classification information determination unit 119 adds the values of the rows of morphemes specified for each column of the first category matrix to calculate the total value. The first classification information determination unit 119 determines the largest value among the calculated total values as the first classification information.

  The case where the character string information "salt ramen" is given will be described based on FIG. The morphological analysis unit 111 decomposes “salt ramen” into morpheme “salt” and “ramen”. The first classification information determination unit 119 specifies the rows related to the morpheme “salt” and “ramen” from the first category matrix based on the morpheme. In the case of morpheme "salt", the proportion of food is "0.42", the proportion of beverage is "0.33", the proportion of books is "0.01", the proportion of stationery is "0.02", the proportion of daily necessities Is "0.04". In the case of the morpheme "ramen", the proportion of food is "0.48", the proportion of beverages is "0.22", the proportion of books is "0.08", the proportion of stationery is "0.05", the proportion of daily necessities Is "0.06". Therefore, the first classification information determination unit 119 calculates the sum for each column. According to FIG. 7, since the value of the food is “0.90”, which is the maximum, the first classification information determination unit 119 determines the first classification information of the character string information “salt ramen” as “food”.

  Next, the case where the first classification information determination unit 119 determines the first classification information based on the first category matrix based on morpheme pairs will be described. The first classification information determination unit 119 identifies, from the first category matrix, rows of morpheme pairs determined by the morpheme pair determination unit 113. The first classification information determination unit 119 determines the largest value among the values of each column of the identified row as the first classification information.

  Based on FIG. 8, the case where the character string information "A company salt noodles" is given will be described. The morphological analysis unit 111 breaks down the "A company salt noodle" into the morpheme "A company", "salt" and "noodle". The morpheme pair determining unit 113 determines the morpheme pair to be “company A” and “noodle”. The first classification information determination unit 119 specifies the rows related to the morpheme pairs “Company A” and “Noodle” from the first category matrix based on the morpheme pairs. In the case of the morpheme pair “Company A” and “noodle”, the ratio of food is “0.97”, the ratio of beverage is “0.01”, the ratio of books is “0.01”, and the ratio of stationery is “0. 00 ", the ratio of daily necessities is" 0.00 ".... Therefore, according to FIG. 8, the value of the food is “0.97”, which is the maximum. Therefore, the first classification information determination unit 119 determines that the first classification information of the character string information “A company salt noodle” is “food”. Decide on.

  The first classification information determination unit 119 may be configured to prioritize the determination result of the morpheme pair in determining the first classification information, or may be configured to prioritize the determination result of the morpheme. Good. For example, when the largest value among the values calculated by the first category matrix of the morpheme pair is the first threshold value or more, the first classification information determination unit 119 determines the first classification information determined by the morpheme pair as the first classification information. You may decide. The first threshold may be, for example, 0.95 or any value. In addition, when the largest value among the values calculated by the first category matrix based on the morpheme pair is greater than or equal to the second threshold and less than the first threshold, the first classification information determination unit 119 determines the first category matrix based on the morpheme. A predetermined numerical value may be added to the first classification information determined by the morpheme pair among the calculation results of (4). The predetermined numerical value may be, for example, 0.1, may be a value calculated by the first category matrix based on morpheme pairs, or may be a value calculated by the first category matrix of morpheme pairs It may be a calculated value after the calculation. The second threshold may be, for example, 0.66 or any value. The first threshold and the second threshold may be designated by the user in advance.

  The second classification information determination unit 120 determines second classification information for the character string information based on the morphemes or morpheme pairs and the second category matrix stored in the category matrix storage unit 107. First, the case where the second classification information determination unit 120 determines the second classification information based on the second category matrix based on the morpheme will be described. The second classification information determination unit 120 specifies the rows of morphemes analyzed by the morpheme analysis unit 111 from the second category matrix. The second classification information determination unit 120 adds the values of the rows of morphemes specified for each column of the second category matrix to calculate a total value. The second classification information determination unit 120 determines the largest value among the calculated total values as the second classification information.

  The case where the character string information "salt ramen" is given will be described based on FIG. The morphological analysis unit 111 decomposes “salt ramen” into morpheme “salt” and “ramen”. The second classification information determination unit 120 identifies the rows related to the morpheme “salt” and “ramen” from the second category matrix based on the morpheme. In the case of morpheme "salt", the value of rice is "2", the value of noodles is "8", the value of bread is "3", the value of meat is "2", the value of vegetables is "3". . In the case of the morpheme "ramen", the value of rice is "1", the value of noodles is "13", the value of bread is "0", the value of meat is "4", the value of vegetables is "5". . Therefore, the second classification information determination unit 120 calculates the sum for each column. According to FIG. 9, since the value of noodles is “21” and is the maximum, the second classification information determination unit 120 determines the second classification information of the character string information “salt ramen” as “noodles”.

  Next, the case where the second classification information determination unit 120 determines the second classification information based on the second category matrix based on morpheme pairs will be described. The second classification information determination unit 120 specifies the rows of the morpheme pair determined by the morpheme pair determination unit 113 from the second category matrix. The second classification information determination unit 120 determines the largest value among the values of each column of the identified row as the second classification information.

  The case where the character string information "A company salt noodles" is given will be described based on FIG. The morpheme pairs of the character string information "A company salt noodle" are "A company" and "noodle". The second classification information determination unit 120 specifies the rows related to the morpheme pairs “Company A” and “Noodle” from the second category matrix based on the morpheme pairs. In the case of the morpheme pair “Company A” and “noodle”, the value of rice is “0”, the value of noodles is “9”, the value of bread is “0”, the value of meat is “0”, the value of vegetables is The value is "0". Therefore, according to FIG. 10, since the value of noodles is “9” and is the maximum, the second classification information determination unit 120 determines the second classification information of the character string information “A company salt noodles” as “noodles”. Do.

  In addition, when determining the second classification information, the second classification information determination unit 120 may be configured to prioritize the determination result of the morpheme pair, or may be configured to prioritize the determination result of the morpheme. Good. For example, if the second classification information determination unit 120 determines that the ratio of the second classification information having the maximum value in the second category matrix of morpheme pairs has a value equal to or greater than the third threshold, the second classification information determination unit 120 determines the morpheme pair. The second classification information may be determined. The third threshold may be, for example, 0.95, or any value. The ratio of the second classification information having the largest value is a value obtained by dividing the value of the second classification information having the largest value by the sum of the values of the respective second classification information. In addition, the second classification information determination unit 120 is configured to use the morpheme pair when the ratio of the second classification information having the largest value is greater than or equal to the third threshold and less than the fourth threshold in the second category matrix of morpheme pairs. A predetermined numerical value may be added to the calculation result of the second category matrix of morphemes with respect to the determined second classification information. The predetermined numerical value may be, for example, 100, may be a value calculated by the second category matrix of morpheme pairs, or is numerically calculated on the value calculated by the second category matrix of morpheme pairs It may be a calculated value. The fourth threshold may be, for example, 0.66 or any value. The third threshold and the fourth threshold may be designated by the user in advance.

  The master information generation unit 121 generates a master information record. The master information generation unit 121 generates a master information record based on the determined first classification information and second classification information and the acquired character string information. The master information generation unit 121 records a master information record in a master information table.

  FIG. 11 is a flowchart showing the flow of the process of generating a category matrix of the information processing apparatus 100 according to the first embodiment. The data acquisition unit 110 acquires training data from the training data storage unit 104 (step S101). The morphological analysis unit 111 performs morphological analysis on character string information (step S102). When the decomposed morpheme is stored in the pre-substitution column of the substitution data table, the morpheme substitution unit 112 substitutes the morpheme with the morpheme of the post-substitution column (step S103). The morphological analysis unit 111 determines the part of speech of the morpheme. The morphological analysis unit 111 deletes the morpheme determined as the “symbol” (step S104). The morpheme pair determining unit 113 determines a combination of morphemes positioned at the beginning and the end of the character string information as morpheme pairs among the morphemes (step S105).

  The first category matrix generation unit 116 generates a first category matrix based on the morpheme and the morpheme pair (step S106). The first category matrix generation unit 116 records the generated first category matrix in the category matrix storage unit 107 (step S107). The second category matrix generation unit 117 generates a second category matrix based on the morpheme and the morpheme pair and the second classification information (step S108). The second category matrix generation unit 117 records the generated second category matrix in the category matrix storage unit 107 (step S109).

  FIG. 12 is a flow chart showing the flow of classification processing of the information processing apparatus 100 of the first embodiment. The data acquisition unit 110 acquires character string information via the communication unit 101 or the input unit 102 (step S201). The master information search unit 118 determines whether character string information is included in a predetermined column of the master information table (step S202). The predetermined column is, for example, a trade name. When character string information is included (step S202: YES), the information processing apparatus 100 ends the process. When character string information is not included (step S202: NO), the classification unit 115 determines classification information (step S203).

  13 and 14 are flowcharts showing the flow of the process of determining classification information of the information processing apparatus 100 according to the first embodiment. The data acquisition unit 110 acquires character string information via the communication unit 101 or the input unit 102 (step S301). The morphological analysis unit 111 performs morphological analysis on character string information (step S302). When the decomposed morpheme is stored in the pre-substitution column of the substitution data table, the morpheme substitution unit 112 substitutes the morpheme with the morpheme of the post-substitution column (step S303). The morphological analysis unit 111 determines the part of speech of the morpheme. The morpheme analysis unit 111 deletes the morpheme determined as the “symbol” (step S304). The morpheme pair determining unit 113 determines a combination of morphemes positioned at the beginning and the end of the character string information as morpheme pairs among the morphemes (step S305).

  The first classification information determination unit 119 acquires a first category matrix based on morphemes and morpheme pairs (step S306). The first classification information determination unit 119 identifies rows of morphemes analyzed by the morpheme analysis unit 111 from the acquired first category matrix, and calculates a total value of values of each row for each column (step S307). The first classification information determination unit 119 identifies the row of morpheme pairs determined by the morpheme pair determination unit 113 from the first category matrix generated based on the morpheme pairs, and the maximum value among the values of each column is determined. Identify.

  The first classification information determination unit 119 determines whether or not the largest value among the values calculated by the first category matrix of morpheme pairs is a value equal to or more than the first threshold (step S308). In the flowchart, the first threshold is 0.95. If the value is 0.95 or more (step S208: YES), the first classification information determination unit 119 determines, among the values held in the first category matrix of morpheme pairs, one of the rows of the identified morpheme pairs The first classification information having a large value is determined as the first classification information of the character string information, and the process proceeds to step S313 (step S309). If the value is not 0.95 or more (step S308: NO), the first classification information determination unit 119 determines, among the values held in the first category matrix of morpheme pairs, the largest of the rows of the identified morpheme pairs It is determined whether the value is equal to or greater than the second threshold (step S310). In the flowchart, the second threshold is 0.66. If the value is equal to or greater than 0.66 (step S310: YES), the first classification information determination unit 119 selects the first classification information determined by the morpheme pair among the values calculated by the first category matrix of morphemes. Then, a predetermined numerical value is added (step S311). The first classification information determination unit 119 determines the largest value among the calculated total values as the first classification information (step S312).

  The second classification information determination unit 120 determines second classification information based on the morpheme and the first classification information determined as the morpheme pair. The second classification information determination unit 120 acquires a second category matrix based on the morpheme and the morpheme pair and the second classification information (step S313). The second classification information determination unit 120 identifies the rows of the morphemes analyzed by the morpheme analysis unit 111 from the second category matrix generated based on the morphemes, and calculates the total value of the values for each column (step S314) . The second classification information determination unit 120 specifies the rows of the morpheme pairs determined by the morpheme pair determination unit 113 from the second category matrix generated based on the morpheme pairs, and the maximum value among the values of each column. Identify.

  The second classification information determination unit 120 determines whether or not the ratio of the second classification information having the largest value in the second category matrix of morpheme pairs is a value equal to or more than the third threshold (step S315). . In the flowchart, the third threshold is 0.95. If the value is 0.95 or more (step S315: YES), the second classification information determination unit 120 selects the second classification information having the largest value among the values calculated in the second category matrix of morpheme pairs. The second classification information of the character string information is determined, and the process proceeds to step S320 (step S316). If the value is not 0.95 or more (step S315: NO), the second classification information determination unit 120 determines that the largest value among the values calculated in the second category matrix of morpheme pairs is the value equal to or more than the fourth threshold. It is determined whether or not (step S317). In the flowchart, the fourth threshold is 0.66. If the value is equal to or greater than 0.66 (step S317: YES), the second classification information determination unit 120 determines the second classification information determined by the morpheme pair among the calculation results of the second category matrix of morphemes. A predetermined numerical value is added (step S318). The second classification information determination unit 120 determines, as the second classification information of the character string information, second classification information having the largest value among the calculated total values (step S319).

  The master information generation unit 121 generates a master information record based on the determined first classification information and second classification information and the acquired character string information (step S320). The master information generation unit 121 records the master information record in the master information table (step S321).

  In the information processing apparatus 100 configured as described above, the first category matrix generation unit 116 and the second category matrix generation unit 117 generate the first category matrix and the second category matrix based on the training data. The morphological analysis unit 111 decomposes the given character string information into morphemes. The morpheme pair determination unit 113 determines the morpheme at the beginning of the character string information and the morpheme at the end as a morpheme pair. The first classification information determination unit 119 determines first classification information indicating the feature of the character string information based on the morpheme and the morpheme pair and the first category matrix. Furthermore, the second classification information determination unit 120 determines second classification information associated with the first classification information based on the morpheme and the morpheme pair, the determined first classification information, and the second category matrix. . Therefore, it is possible to classify information more easily.

Second Embodiment
Next, an information processing system 1 according to the second embodiment will be described. FIG. 15 is a system configuration diagram showing a system configuration of the information processing system 1 of the second embodiment. The information processing system 1 is provided in the network 400. The information processing system 1 adds information indicating the feature of any character string information to the input character string information. The information processing system 1 includes an information processing device 100 a and a terminal device 300 which are communicably connected to each other via a network 400. The network 400 may be established by any network. For example, the network 400 may be configured by the Internet.

  The information processing apparatus 100a according to the second embodiment is different from the first embodiment in that the information processing apparatus 100a includes a control unit 109a instead of the control unit 109 and includes a classification unit 115a instead of the classification unit 115. The configuration is the same. The differences from the first embodiment will be described below.

  The control unit 109a controls the operation of each unit of the information processing apparatus 100a. The control unit 109a is executed by, for example, an apparatus including a processor and a memory. The control unit 109a functions as a data acquisition unit 110, a morpheme analysis unit 111, a morpheme replacement unit 112, a morpheme pair determination unit 113, a learning unit 114, and a classification unit 115a by executing a classification program or a learning program.

  The data acquisition unit 110a acquires character string information. The data acquisition unit 110 a acquires character string information from the terminal device 300. The data acquisition unit 110a outputs the acquired character string information to the classification unit 115a. The data acquisition unit 110a may record the acquired character string information in the training data storage unit 104.

  The classification unit 115a functions as a master information search unit 118a, a first classification information determination unit 119, a second classification information determination unit 120, and a master information generation unit 121 by executing a classification program.

  The master information search unit 118a determines whether character string information is included in a predetermined column included in the master information table in the same process as the master information search unit 118 of the first embodiment. Based on the master information table stored in master information storage unit 108, master information search unit 118a determines whether or not character string information is included. When character string information is included in the master information table, the master information searching unit 118a transmits, to the terminal device 300, the first classification information and the second classification information associated with the character string information. When character string information is not included in the master information table, the classification unit 115a executes classification information determination processing. The master information search unit 118a transmits, to the terminal device 300, the first classification information and the second classification information determined by the execution result of the classification information determination process.

  The terminal device 300 includes a processor, a memory, an auxiliary storage device, and the like connected by a bus, and functions as a device including the communication unit 301, the input unit 302, the display unit 303, and the control unit 304 by executing a classification request program. Do. Note that all or part of the functions of the terminal device 300 may be realized using hardware such as an ASIC, a PLD, or an FPGA. The classification program or the learning program program may be recorded on a computer readable recording medium. The computer readable recording medium is, for example, a portable medium such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, or a storage device such as a hard disk built in a computer system. The classification program or the learning program program may be transmitted via a telecommunication line.

  The communication unit 301 is a network interface. The communication unit 301 communicates with an external device via the network 400. The external device may be, for example, a device such as the information processing device 100a. The communication unit 301 may communicate by a communication method such as wireless LAN, wired LAN, Bluetooth, or LTE, for example.

  The input unit 302 is configured using an input device such as a touch panel, a mouse, and a keyboard. The input unit 302 may be an interface for connecting an input device to the terminal device 300. In this case, the input unit 302 generates input data (for example, instruction information indicating an instruction to the terminal device 300) from an input signal input to the input device, and inputs the input data to the terminal device 300.

  The display unit 303 is an output device such as a CRT display, a liquid crystal display, or an organic EL display. The display unit 303 may be an interface for connecting an output device to the terminal device 300. In this case, the display unit 303 generates a video signal from the video data and outputs the video signal to a video output device connected to itself.

  The control unit 304 controls the operation of each unit of the terminal device 300. The control unit 304 is executed by, for example, an apparatus including a processor and a memory. The control unit 304 functions as an apparatus including the classification request unit 305 by executing the classification request program.

  The classification request unit 305 receives character string information from the user via the input unit 302. The classification request unit 305 transmits the received character string information to the information processing apparatus 100a. The classification request unit 305 receives the first classification information and the second classification information from the information processing apparatus 100a. The classification request unit 305 causes the display unit 303 to display the received first classification information and second classification information.

<Modification>
In the present embodiment, the morpheme pair is described as being the morpheme at the beginning and the end of the character string information, but the morpheme pair is not limited to the beginning and the end of the character string information. For example, the morpheme pair determining unit 113 may determine the first and second morphemes from the beginning of the character string information as a morpheme pair.

  The morpheme pair determination unit 113 specifies a morpheme that most represents a feature of the character string information among morphemes in which the character string information is decomposed. The morpheme that most represents the feature of the character string information is a morpheme whose meaning is similar to that of the character string information by comparing the character string information and the morpheme. The morpheme pair determination unit 113 may use a known method in identifying morphemes whose meanings are similar to character string information. The morpheme pair determining unit 113 identifies another morpheme whose meaning is similar to the identified morpheme. The morpheme pair determining unit 113 determines the identified morpheme and another morpheme whose meaning is similar to the identified morpheme as a morpheme pair. The morpheme pair determining unit 113 may use a known method in identifying another morpheme having a meaning similar to that of the morpheme.

  In the present embodiment, the training data is stored in the training data storage unit 104, but the training data may be input from the outside via the communication unit 101 as needed, or the user via the input unit 102. You may accept from

  In the present embodiment, the classification process and the learning process are described as being performed in the same information processing apparatus 100 or 100 a. However, the present invention is not limited to this. For example, the classification device that performs classification processing and the learning device that performs learning processing may be respectively configured, or may be configured on cloud computing.

  The embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes design and the like within the scope of the present invention.

  The present invention is applicable to the classification of goods.

100 ... information processing apparatus, 101 ... communication unit, 102 ... input unit, 103 ... display unit, 104 ... training data storage unit, 105 ... replacement data storage unit, 106 ... classification information storage unit, 107 ... category matrix storage unit, 108 ... Master information storage unit, 109 ... Control unit, 110 ... Data acquisition unit, 111 ... Morphological analysis unit, 112 ... Morpheme replacement unit, 113 ... Morpheme pair determination unit, 114 ... Learning unit, 115 ... Classification unit, 116 ... 1st Category matrix generation unit, 117: second category matrix generation unit, 118: master information search unit, 119: first classification information decision unit, 120: second classification information decision unit, 121: master information generation unit, 1: information processing System, 100a ... information processing apparatus, 109a ... control unit, 110a ... data acquisition unit, 115a ... classification unit, 118a ... ma Data information search section, 300 ... terminal apparatus, 301 ... communication unit, 302 ... input unit, 303 ... display unit, 304 ... controller, 305 ... classification request unit, 400 ... network

One aspect of the present invention, the morpheme analysis unit for dividing the character string to be classified into morphemes, the first classification information given as the information indicating the feature of the learning target string to a string of the learned learning classification based on a plurality of training data to each morpheme included in the character string of the learned by the predetermined machine learning, to the first classification information the classification target string is one that contains a target string Gets the previous Kigaku learning result data a percentage representing how is the learning result data storage unit that records including learning result data, before Kigaku learning result data, the morphological analysis from the string of the classification target And a first classification information determination unit that determines the first classification information of the character string to be classified based on the morphemes divided by a unit.

According to one aspect of the present invention, there is provided a morphological analysis unit that divides a classification target name into morphemes, first classification information added to the learning target name as information indicating characteristics of the learning target name, and the learning target character By performing predetermined machine learning based on a plurality of training data including a column, it is determined which first classification information the name of the classification target is classified for every morpheme included in the character string of the learning target Learning result data from the learning result data storage unit that records the learning result data including the frequency to be expressed, based on the learning result data and the morpheme divided by the morpheme analysis unit from the name of the classification target Te, a first classification information determination unit configured to determine the first classification information of the classified name, positioned at the beginning and end of the plurality of the morphemes derived from the name of the classification target One of comprising a morpheme pair determination unit that determines the morpheme pair including morphological, and the learning result data of the plurality included in the learning object name by a predetermined machine learning based on the plurality of training data Learning result data including a frequency indicating which first classification information the name of the classification target is to be classified for every morpheme pair including two morphemes positioned at the beginning and the end of morphemes, the first A classification information determination unit includes the morpheme pair determined by the morpheme pair determination unit based on the name of the classification target, the morpheme divided by the morpheme analysis unit from the classification target name, and the learning result data And a classification device that determines the first classification information of the name of the classification target .

According to one aspect of the present invention, there is provided a morphological analysis unit that divides a character string to be classified into morphemes, first classification information added to the character string to be learned as information indicating a feature of the character string to be learned, and the learning The character string to be classified is classified into any first classification information for each morpheme included in the character string to be learned by performing predetermined machine learning based on a plurality of training data including the character string to be targeted The learning result data is stored from the learning result data storage unit that records learning result data including a frequency indicating whether the learning result data is stored, and the learning result data is divided by the morpheme analysis unit from the classification result character string. A first classification information determination unit that determines the first classification information of the character string to be classified based on a morpheme; and the training data includes the attached first classification information; 1 classification Data including second classification information which is a subclass of the information, and the learning result data is obtained by performing predetermined machine learning based on the plurality of training data, and each morpheme contained in the character string of the learning target in a learning result data including the frequency indicating whether the classification target string is classified into one of the second classification information, before Kigaku learning result data and, from said character string of the classified morphological analysis unit based on the above morphological divided by, before Symbol further comprising a second classification information determining unit for determining a second classification information is a classification device.

According to one aspect of the present invention, there is provided a morphological analysis unit that divides a character string to be classified into morphemes, first classification information added to the character string to be learned as information indicating a feature of the character string to be learned, and the learning The character string to be classified is classified into any first classification information for each morpheme included in the character string to be learned by performing predetermined machine learning based on a plurality of training data including the character string to be targeted The learning result data is stored from the learning result data storage unit that records learning result data including a frequency indicating whether the learning result data is stored, and the learning result data is divided by the morpheme analysis unit from the classification result character string. morpheme, based on a first classification information determination unit configured to determine the first classification information string of the classification target, the morphemes are divided from the character string of the classified by the morphological analysis unit Based on a predetermined condition comprises a morpheme replacement unit which replaces the other morphemes, and a classifier.

According to one aspect of the present invention, there is provided a morphological analysis unit that divides a classification target name into morphemes, first classification information added to the name of the learning target as information indicating characteristics of the learning target name, and a name of the learning target And performing predetermined machine learning on the basis of a plurality of training data including, to indicate which first classification information the name of the classification target is classified for each morpheme contained in the character string of the learning target The learning result data is acquired from a learning result data storage unit that records learning result data including a frequency, and based on the learning result data and the morpheme divided by the morpheme analysis unit from the name of the classification target a first classification information determination unit configured to determine the first classification information of the classified name, most meaningful and the classification target name among the plurality of morphemes obtained from the name of the classification target There was determined the morphemes, and a morpheme pair determination unit that determines the morpheme pair comprising the other morphological meaning is closer to the morphemes identified, wherein the learning result data of predetermined based on said plurality of training data machine The name of the classification target is classified into any first classification information for each morpheme pair including two morphemes positioned at the beginning and the end among a plurality of morphemes included in the name of the learning object by learning or a learning result data including the frequency with which represents a first classification information determining unit includes: the learning result data, and the morpheme pair determined by the morpheme pair determination unit based on the name of the classification target, the It is a classification apparatus which determines said 1st classification information of the name of said classification object based on .

One aspect of the present invention, the morpheme analysis unit for dividing the character string to be learned into morphemes, the first classification information given as the information indicating the feature of the learning target string to a string of the learned learning By performing predetermined machine learning based on a plurality of training data including the target character string, the character string to be classified is classified into any first classification information for each morpheme included in the character string to be learned. comprising a learning result data generation unit for generating including learning result data the percentage representing the Luke, a, is a learning device.

One aspect of the present invention, the classification device, the morphological analysis step of dividing the character string to be classified into morphemes, the classification device, is applied to the string of the learned as information indicating a characteristic of the learning target string By performing predetermined machine learning based on a plurality of training data including the first classification information and the character string to be learned, the character string to be classified is for each morpheme included in the character string to be learned. Gets the previous Kigaku learning result data the percentage representing the either fall into one of the first classification information from the learning result data storage unit that records including learning result data, before Kigaku learning result data, the classification It said morphemes from target string is divided in the morphological analysis step, based on, having a first classification information determining step of determining the first classification information string of the classified, the classification method is there.

One aspect of the present invention, the learning device, the morphological analysis step of dividing the character string to be learned into morphemes, the learning device is assigned to a string of the learned as information indicating a characteristic of the learning target string to each morpheme included in the character string of the learned by the predetermined machine learning based on a plurality of training data including a first classification information and the character string of the learned, the string to be classified is either a It has a learning result data generating step of a percentage representing how is classified into the first classification information to generate including learning result data, and a learning method.

Claims (10)

A morphological analysis unit that divides a character string into morphemes;
By performing predetermined machine learning based on a plurality of training data including a character string and the classification result of the character string, the frequency in which the first classification information representing the feature of the character string has been given to the character string so far Acquiring the first learning result data from the learning result data storage unit that records first learning result data including the first learning result data, and based on the first learning result data and the morpheme, the first classification information of the character string A first classification information determination unit to determine
A sorting device comprising: And a morpheme pair determining unit configured to determine morpheme pairs including any two of the morphemes based on a predetermined condition,
The first classification information determination unit performs predetermined machine learning based on the training data, so that the first classification information becomes a character string based on morphemes included in morpheme pairs of the training data. The second learning result data is acquired from the learning result data storage unit that records the second learning result data including the number of times assigned, and the character string is acquired based on the second learning result data and the morpheme pair. The classification device according to claim 1, wherein the first classification information is determined. By performing predetermined machine learning based on the training data, second classification information representing the feature of the character string is determined among the plurality of second classification information associated with the determined first classification information. Acquiring the third learning result data from the learning result data storage unit that records third learning result data for the second learning, and determining the second classification information based on the third learning result data and the morpheme 2 further comprising a classification information determination unit,
The classification device according to claim 1.   The classification device according to claim 1, further comprising a morpheme substitution unit that substitutes the morpheme with another morpheme based on a predetermined condition. The morpheme pair determining unit determines a morpheme closest in meaning to the character string among the morphemes, and determines a morpheme pair including the identified morpheme and another morpheme whose meaning is similar,
The first classification information determination unit performs predetermined machine learning based on the training data, so that the first classification information becomes a character string based on morphemes included in morpheme pairs of the training data. The second learning result data is acquired from the learning result data storage unit which records the second learning result data including the number of times assigned, and the character string is acquired based on the second learning result data and the morpheme pair. The classification apparatus according to claim 1, wherein the first classification information of is determined. A morphological analysis unit that divides a character string into morphemes;
By acquiring a plurality of training data including a character string and the classification result of the character string and performing predetermined machine learning, the first classification information representing the feature of the character string has been given to the character string so far A learning result data generation unit that generates first learning result data including a frequency;
, A learning device. A morphological analysis step in which the classification device divides the character string into morphemes;
The classification device performs predetermined machine learning based on a plurality of training data including a character string and the classification result of the character string, whereby the first classification information representing the feature of the character string is given to the character string so far The first learning result data is acquired from a learning result data storage unit that records first learning result data including the frequency of occurrence, and the first learning result data is acquired based on the first learning result data and the morpheme. 1 first classification information determination step of determining classification information;
Have a classification method. A morphological analysis step in which the learning device divides the character string into morphemes;
The learning device acquires a plurality of training data including a character string and the classification result of the character string, and performs predetermined machine learning, so that the first classification information representing the feature of the character string becomes the character string so far A learning result data generation step of generating first learning result data including the assigned frequency;
Have learning methods.   A computer program for causing a computer to function as the classification device according to any one of claims 1 to 5.   A computer program for causing a computer to function as the learning device according to claim 6.

Images