Information processing method, information processing system, and program

Abstract

To accurately recognize a collection of text data in irregular document data.SOLUTION: An information processing method causes a computer to: acquire irregular document data including a plurality of pieces of character information; convert the irregular document data into text data in which each of the plurality of pieces of character information is arranged on each line; and add classification information, to the text data, for classifying each line in the irregular document data into a predetermined segment using a model obtained by machine learning based on the relationship between characters in the same line and the relationship between adjacent lines.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an information processing method, an information processing system, and a program. [Background technology]

[0002] In recruitment operations such as new graduate and mid-career hiring, hiring managers read PDF (Portable Document Format) files (data) of resumes and career histories to screen applicants. At this time, the hiring manager manually selects applicants by extracting and comparing the qualifications and experience possessed by each applicant. As a result, a large amount of cost is spent on recruitment operations, and there are hopes for the creation of a database of resume information. Patent Document 1 discloses the creation of a database of resumes and the extraction of each piece of information. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Publication No. 2017-097498 Summary of the Invention [Problem to be solved by the invention]

[0004] On the other hand, if resumes and careers are PDF data, they can be converted into text by processing such as OCR (Optical Character Recognition), but visual information such as lines are deleted. For this reason, a human being can recognize each collection of text data (also called segments, classifications, attributes, groupings, etc.) such as educational background, career, and qualifications held with only text data, but a computer cannot recognize each collection of text data when performing calculation processing. In addition, resumes and careers come in various formats (also called irregular document data), and the order and notation of educational background, career, and qualifications held may differ. For this reason, even if only character string processing is performed on PDF data of resumes and careers, it is not possible to accurately recognize the collection of text data.

[0005] One object of the present invention is to accurately recognize chunks of text data in irregular document data. [Means for solving the problem]

[0006] According to one embodiment of the present invention, there is provided an information processing method in which a computer acquires irregular document data in which multiple pieces of character information are arranged in segments, converts the irregular document data into text data in which each piece of the multiple pieces of character information is arranged in each line based on its relative positional relationship, and generates classification information for classifying each line in the text data into a predetermined segment using a model obtained by machine learning the relationship between character information in the same line, character information arranged in adjacent lines, and segments corresponding to the text data of each line.

[0007] In the above information processing method, the machine learning may be supervised machine learning that uses training data including first identification information that corresponds to and can be identified by the classification information and second identification information that can be identified for each character to learn relationships between character information in the same line, character information arranged in adjacent lines, and segments corresponding to text data of each line in the text data.

[0008] In the above information processing method, the classification information may include second classification information corresponding to a second segment indicating a collection of text data in one segment.

[0009] In the above information processing method, in the model, an embedding vector may be generated based on the second identification information by using an embedding vector data table configured as a vector corresponding to each character.

[0010] In the above information processing method, the model may use the embedding vector to generate a first feature vector between characters in the same line and a second feature vector between adjacent lines, and generate an estimate for estimating the classification information based on the first feature vector and the second feature vector.

[0011] In the above information processing method, the irregular document data may be resume data, and the resume data may include history information associated with each user.

[0012] According to one embodiment of the present invention, there is provided an information processing system including a control unit that acquires irregular document data in which multiple pieces of character information are arranged in segments, converts the irregular document data into text data in which each piece of the multiple pieces of character information is arranged in each line based on its relative positional relationship, and generates classification information for classifying each line in the text data into a predetermined segment using a model obtained by machine learning the relationship between character information in the same line, character information arranged in adjacent lines, and segments corresponding to the text data of each line in the text data.

[0013] In the above information processing system, the machine learning may be supervised machine learning that uses training data including first identification information that corresponds to and can be identified by the classification information and second identification information that can be identified for each character, to learn the relationship between character information within the same line, character information arranged in adjacent lines, and segments corresponding to the text data of each line in the text data.

[0014] In the above information processing system, the classification information may include second classification information corresponding to a second segment indicating a collection of text data in one segment.

[0015] In the above information processing system, in the model, an embedding vector may be generated based on the second identification information using an embedding vector data table configured as a vector corresponding to each character.

[0016] In the above information processing system, the model may use the embedding vector to generate a first feature vector between characters in the same line and a second feature vector between adjacent lines, and generate an estimate for estimating the classification information based on the first feature vector and the second feature vector.

[0017] In the above information processing system, the irregular document data may be resume data, and the resume data may include history information associated with each user.

[0018] According to one embodiment of the present invention, a program is provided that causes a computer to acquire irregular document data in which multiple pieces of character information are arranged in segments, convert the irregular document data into text data in which each piece of the multiple pieces of character information is arranged in each line based on its relative positional relationship, and generate classification information for classifying each line of the text data into a predetermined segment using a model obtained by machine learning the relationship between character information in the same line, character information arranged in adjacent lines, and segments corresponding to the text data of each line. Effect of the Invention

[0019] According to the present invention, it is possible to accurately recognize chunks of text data in irregular document data. [Brief description of the drawings]

[0020] [Figure 1] 1 is a diagram illustrating a hardware configuration of an information processing system according to an embodiment of the present invention. [Diagram 2] FIG. 1 shows a functional block diagram of each component of an information processing system according to an embodiment of the present invention. [Diagram 3] FIG. 1 is a flow diagram of an information processing system according to an embodiment of the present invention. [Figure 4] 4 is an example of text data according to an embodiment of the present invention. [Diagram 5] 1 is a flowchart of teacher data generation according to one embodiment of the present invention. [Figure 6] 1 is an example of annotated text data for teacher data according to one embodiment of the present invention. [Figure 7] 1 is an example of text data for teaching data that has been chunked according to an embodiment of the present invention. [Figure 8] 13 is a diagram illustrating an example of a case where numbers are assigned to text data according to an embodiment of the present invention. [Figure 9]13 is a diagram illustrating an example of a case where numbers are assigned to identification information according to an embodiment of the present invention. [Figure 10] FIG. 1 is a schematic diagram of a machine learning model according to an embodiment of the present invention. [Figure 11] 1 is a flow chart of a machine learning model according to an embodiment of the present invention. [Figure 12] FIG. 1 is a schematic diagram of LSTM calculation processing. [Figure 13] 1 shows text data to which label information has been added according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] Hereinafter, the embodiments of the present invention will be described with reference to the drawings. However, the present invention can be implemented in many different forms, and should not be interpreted as being limited to the description of the embodiments exemplified below. The drawings may be schematic in order to make the explanation clearer, but they are merely examples and do not limit the interpretation of the present invention.

[0022] Furthermore, the letters "first" and "second" attached to each element are convenient labels used to distinguish each element, and have no further meaning unless otherwise specified. In the drawings referred to in this embodiment, the same parts or parts having similar functions are given the same or similar symbols (symbols simply consisting of a number with A, B or 1, 2, etc. attached thereto), and repeated explanations may be omitted. Also, some of the configurations may be omitted from the drawings. In addition, if a person having ordinary knowledge in the field to which the present invention belongs can recognize the above, no special explanation will be given.

[0023] First Embodiment An information processing system according to a first embodiment of the present invention will be described in detail with reference to the drawings.

[0024] (1-1. Overall configuration of information processing system) FIG. 1 is a diagram for explaining a hardware configuration of an information processing system 1 in a first embodiment of the present invention. As shown in FIG. 1, the information processing system 1 includes an information processing server 10 and a user communication terminal 30. The information processing server 10 functions as a database for information processing and an application server. The user communication terminal 30 is a terminal of a user (in this example, a person in charge of recruiting) who uses irregular document data (in this example, a resume) that has been converted into PDF. In this embodiment, for convenience of explanation, a case where one user communication terminal 30 is used is shown, but multiple user communication terminals 30 may be used as appropriate.

[0025] In the information processing system 1, the user communication terminal 30 transmits irregular document data (resume data in this example) to the information processing server 10 via the network NW. The information processing server 10 converts the irregular document data into text data, and then uses a machine learning model to generate text data to which label information (also referred to as "classification information") is assigned for each line.

[0026] The information processing server 10 includes a control unit 11, a storage unit 12, a communication unit 13, and a display unit 14.

[0027] The control unit 11 is a computer, and uses a CPU (Central Processing Unit), ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), or other arithmetic processing circuit to control processing based on instructions defined in information processing control software (program) described later. In addition, a user interface for executing the information processing control program may be provided on the display unit 14 by instructions from the control unit 11.

[0028] The storage unit 12 may be a semiconductor memory such as an SSD (Solid State Drive), a magnetic recording medium (magnetic tape, magnetic disk, etc.), an optical recording medium, a magneto-optical recording medium, or a memorizable element that is a storage medium. The storage unit 12 functions as a database that stores information used in the information processing control program. The storage unit 12 may be provided in a server different from the information processing server 10 as appropriate, and may function as a database.

[0029] The communication unit 13 has a transceiver and communicates with the user communication terminal 30 via the network NW. The communication between the communication unit 13 and the user communication terminal 30 is performed using the Internet (specifically, SSL / TLS or VPN).

[0030] The user communication terminal 30 has a display unit 31, a control unit 32, a storage unit 33, an operation unit 34, and a communication unit 35. The user communication terminal 30 may be a smartphone, a mobile phone (feature phone), a tablet terminal, a notebook PC, a desktop PC, an IoT device (equipment equipped with a power supply mechanism, a communication function, and an information storage mechanism), etc., and is applicable as long as it is capable of communicating with the information processing server 10 via a network.

[0031] The display unit 31 is a display device such as a liquid crystal display or an organic EL display, and the display content is controlled by a signal input from the control unit 32.

[0032] The control unit 32 is a computer and includes a CPU, an ASIC, an FPGA, or other arithmetic processing circuit. The control unit 32 executes a program stored in a storage unit 33 such as a memory based on operations on the display unit 31 and the operation unit 34, and instructs the execution of processing related to the information processing control program stored in the storage unit 12 of the information processing server 10.

[0033] The operation unit 34 includes a controller, a button, or a switch. When the operation unit 34 is moved up, down, left, right, pressed, rotated, or the like, information based on the operation is transmitted to the control unit 32. Note that if the display unit 31 and the operation unit 34 are a display device (touch panel) having a touch sensor, the display unit 31 and the operation unit 34 may be disposed in the same place.

[0034] The communication unit 35 has a function of transmitting and receiving with the information processing server 10. For example, a LAN transceiver (e.g., a Wi-Fi transceiver) is used for the communication unit 35. Note that the transceiver is not limited to a LAN transceiver, and a transceiver for mobile terminal communication (e.g., LTE communication) or a transceiver for short-range wireless communication may be provided. The user communication terminal 30 is connected to the information processing server 10 via a network NW.

[0035] (1-2. Configuration of Information Processing Control Unit 100) FIG. 2 is a functional block diagram of the components of the information processing system 1.

[0036] In the information processing system 1, the information processing server 10 (control unit 11) has an information processing control unit 100 that controls a program that realizes an information processing function. The information processing control unit 100 includes an acquisition unit 110, a conversion unit 120, a machine learning unit 130, a generation unit 140, an addition unit 150, and a transmission unit 160. In addition, the user communication terminal 30 has a transmission unit 310 and a reception unit 330.

[0037] The acquisition unit 110 has a function of performing each acquisition process in the information processing of this embodiment. In this example, the acquisition unit 110 acquires irregular document data (resume data) and various data transmitted from the user communication terminal 30. The acquired information is stored in a text data database (DB) 12a provided in the storage unit 12.

[0038] The conversion unit 120 has a function of converting one piece of data into a different data format. Specifically, the conversion unit 120 has a function of converting irregular document data into text data, a function of converting text data into numeric data, a function of converting numeric data into an embedding vector (feature vector), and a function of converting an estimated value into label information. The conversion unit 120 may also reconvert each piece of data into the original data format.

[0039] The machine learning unit 130 has a function of optimizing a machine learning model (also simply called a model) using teacher data and learning text data.

[0040] The generating unit 140 has a function of generating an estimated value for each piece of text data based on a machine learning model, and a function of generating other information.

[0041] The adding unit 150 has a function of adding generated label information (classification information) to each piece of text data.

[0042] The transmitting unit 160 has a function of transmitting the text data to which the label information has been added, and various types of information, to the user communication terminal 30.

[0043] The transmission unit 310 has a function of transmitting to the information processing server 10 the irregular document data to which label information is to be added, the irregular document data for machine learning, and various other data.

[0044] The receiving unit 330 has a function of receiving text data to which label information has been added, and various types of information from the information processing server.

[0045] (1-3-3. Information processing method) The information processing method according to this embodiment will be described below with reference to a flowchart of FIG 3.

[0046] First, the information processing server 10 acquires irregular document data (step S100). In this example, the irregular document data is resume data converted to PDF. In this embodiment, the resume data includes document data used in job hunting, such as resumes or curriculum vitae. The resume data includes history information showing the history of each user, and multiple pieces of character information are arranged for each segment, such as personal information, job summary, and skills. In this embodiment, the irregular document data refers to a document with no fixed format, or a document with multiple formats. Furthermore, the "character information" in this embodiment includes information that can be independently identified in the irregular data, such as characters, spaces, and symbols.

[0047] Next, the information processing server 10 converts the irregular document data into text data (step S200). In this example, OCR is used as the conversion method. Fig. 4 is an example of text data 500. As shown in Fig. 4, in the text data 500, each of a plurality of pieces of character information is arranged on each line based on a relative positional relationship.

[0048] Next, the information processing server 10 generates and adds classification information for classifying each line in the text data into a predetermined segment based on the machine learning model (step S300). The classification information is information for classifying each line in the irregular document data into a predetermined segment. Specifically, the identification information corresponds to "job summary", "job history", "skills", etc. Furthermore, for lines that are not classified into a segment, the identification information of "○" is added. The machine learning model in this embodiment is obtained by machine learning the relationship between character information in the same line, character information (input information) arranged in adjacent lines, and segments (output information) corresponding to the text data of each line in the text data. Here, the machine learning process and the machine learning model used in this embodiment will be described below.

[0049] [Machine learning processing] The machine learning process in this embodiment is performed after the generation and preprocessing of training data. The configuration of the machine learning process will be described below.

[0050] [Teacher data generation] 5 is a flowchart of teacher data generation. First, the information processing server 10 acquires learning text data (step S3101). The learning text data is data obtained by converting irregular document data, including character information, ruled lines, parentheses, etc., accumulated in the past, into text data by OCR processing. The learning text data is acquired, for example, from the text data database (DB) 12a of the information processing server 10.

[0051] Next, text data to be teacher data (text data for teacher data) is selected from the learning text data (step 3102), and the selected text data for teacher data is annotated (step 3103). FIG. 6 shows an example of annotated teacher data 600. As shown in FIG. 6, in this example, as annotation, label information (also called "classification information") 621 corresponding to a segment of text data 611 in each row of a text column 610 is added to each row of a label column 620 for the text data for teacher data. Specifically, information such as "O", "job summary", "skills", and "work history" is added as label information.

[0052] Next, chunking processing is performed on the text data for teacher data (step 3104). FIG. 7 shows an example of text data for teacher data 601 that has been subjected to chunking processing. In this example, as chunking processing, second classification information 623 corresponding to a segment ("second segment") indicating a collection of text data is added to each piece of label information 621 (also called "first segment"). Specifically, "B_" is added to label information arranged in the first row of the same label information 621. Also, "E_" is added to label information arranged in the last row of the same label information. Furthermore, "I_" is placed on label information arranged in other rows of the same label information. Label information to which the second classification information has been added may be used as label information (classification information).

[0053] Next, a first conversion data table is generated for assigning an identifiable number (also called "second identification information") to each independent character included in the text data for training data (step S3105). FIG. 8 shows an example of assigning numbers to text data. As shown in FIG. 8, a different number is assigned to each character. Specifically, in the case of a character string "presentation", "1" is assigned to "pu", "2" to "re", "3" to "ze", "4" to "n", "5" to "te", "6" to "-", "7" to "shi", and "8" to "yo". The same number ("4") is assigned to duplicate characters ("n" in this case).

[0054] Next, a second conversion data table is generated for assigning identifiable numbers (also called "first identification information") to the label information (classification information) included in the text data for training data (step 3106). FIG. 9 shows an example of assigning numbers to label information. As shown in FIG. 9, a different number is assigned to each piece of label information. Specifically, "1" is assigned to "O", "2" to "B_Job Summary", "3" to "I_Job Summary", "4" to "E_Job Summary", "5" to "B_Skills", "6" to "I_Skills", "7" to "E_Skills", "8" to "B_Work Experience", "9" to "I_Work Experience", and "10" to "E_Work Experience".

[0055] Next, the text portion of the text data for teacher data is converted into numeric data set based on the first conversion data table, and the label information portion is converted into numeric data based on the second conversion data table (step 3107). This completes the generation of teacher data.

[0056] Next, a learning process of a machine learning model (also simply referred to as a "model") is performed using the teacher data and learning text data generated by the above process. The machine learning model used in this embodiment will be described below.

[0057] [Machine learning model] Fig. 10 is a schematic diagram (functional blocks) of a machine learning model. Fig. 11 is a flowchart of the machine learning model. In this embodiment, the process from embedding processing to processing in CRF (Conditional Random Fields) (dashed line portion in Fig. 10) is set as the machine learning model, and the other processes can be set as pre-processing and post-processing.

[0058] First, as a pre-processing, the information processing server 10 converts the text data 700 into digitized text data 702 based on the data stored in the above-mentioned first conversion data table 701 as shown in FIG. 10 (step S3201).

[0059] Next, embedding processing is performed on the converted digitized text data 702 (step S3202). Specifically, the information processing server 10 converts each of the digitized text data 702 into a feature vector. At this time, an embedding vector 704 is generated using an embedding vector data table 703. In this embodiment, each character (number) may be represented in 100 dimensions or more, preferably 200 dimensions or more, and more preferably 256 dimensions or more.

[0060] Next, a feature amount (feature vector) for each line is extracted and obtained (step S3203). At this time, a first LSTM (Long Short Term Memory) calculation process is executed. FIG. 12 is a schematic diagram of the LSTM calculation process. In the LSTM calculation process, an input value Xt, a previous hidden state Ht-1 (feature amount), and a previous output value Ct-1 are input, and an output value Ct and a hidden state Ht (feature amount) 705 are output. The output value Ct corresponds to information indicating the correlation between lines. The hidden state Ht (first feature vector 705) corresponds to feature information indicating the correlation between characters. By the first LSTM calculation process, feature information indicating the correlation between characters in the same line (horizontal direction) can be obtained.

[0061] Next, the feature vector from the beginning to the end of the same line is converted into a second feature vector 706 that takes into account the correlation between the beginning and end (step S3204). At this time, a second LSTM calculation process is executed. The second LSTM calculation process may be performed in the same manner as the first LSTM calculation process. The second LSTM calculation process may obtain feature information between lines.

[0062] Next, an estimated value 707 for estimating label information is generated from the first feature vector output by the first LSTM calculation process and the second feature vector output by the second LSTM calculation process (step S3205). At this time, a calculation process using CRF (Conditional Random Fields) is executed.

[0063] In this embodiment, the feature vector in the embedding process is set to a random value (numerical value) in the early learning stage, but changes to an optimal value by repeating learning. In other words, by repeating the learning process of the machine learning model using the teacher data and learning text data, the parameters in the machine learning model are optimized (the accuracy of the estimated value is improved), and a trained machine learning model is generated.

[0064] Next, the generated estimated value is converted based on the second conversion data table 708 to generate label information (classification information). The above is the information processing method in this embodiment.

[0065] FIG. 13 shows a data set 800 in which label information 820 generated by using the information processing method of this embodiment is added to text 810.

[0066] In this embodiment, by learning the relationships between characters and between lines in text data, it is possible to classify each line into an accurate segment even when new irregular document data is converted into text data.

[0067] In addition, in the conventional case, word segmentation processing and morphological analysis are performed on text data to generate embedding vectors on a word basis, but when correcting word segmentation, a new embedding vector data table must be generated (depending on how the words are segmented). Therefore, it is difficult to apply the fine tuning (additional learning) method, and machine learning must be performed again from the beginning. On the other hand, in the present embodiment, an embedding vector (number) is assigned to each individual character. In the present embodiment, when a new character is added, it is sufficient to add it to the end of the first change data table and the embedding vector table, and the fine tuning (additional learning) method can be easily applied to the learned embedding vector data table, LSTM1, LSTM2, and CRF. Therefore, there is no need to perform machine learning from the beginning, and the learning time for fine tuning (additional learning) can be shortened. In other words, by using this embodiment, the calculation processing in the control unit can be simplified. Therefore, by using this embodiment, the load on the control unit 11 (information processing control unit 100) can be reduced.

[0068] In addition, in this embodiment, by including second classification information corresponding to a second segment indicating a grouping of text data in one segment (label information, classification information), it is possible to more accurately recognize groupings of data in amorphous document data.

[0069] In particular, when resume data is used as an irregular document, text information is arranged across multiple lines in one segment to indicate history information (career information) such as educational background and work history for each user, and characteristic information such as skills. By using this embodiment, it is possible to accurately grasp the collection of text data (career history), and therefore, by using the output information, recruiters can reduce the amount of data processing (time) required in recruiting activities (reducing the workload).

[0070] (Modification) Within the scope of the concept of the present invention, a person skilled in the art may come up with various modifications and alterations, and it is understood that these modifications and alterations also fall within the scope of the present invention. For example, those in which a person skilled in the art appropriately adds, deletes, or modifies components of the above-mentioned embodiments, or combines the embodiments, or adds, omits, or modifies conditions of processes, are also included in the scope of the present invention as long as they include the gist of the present invention.

[0071] In one embodiment of the present invention, resume data is used as the irregular document data, but the present invention is not limited to this. Any irregular document data can be used as appropriate as long as it does not contradict the purpose of the present invention. For example, medical data (medical charts), receipt data, accounting data, and other document data may be used as the irregular document data.

[0072] In the embodiment of the present invention, an example in which information processing is performed by the information processing server 10 has been shown, but the present invention is not limited to this. When the information processing control program is stored in the storage unit 33 of the user communication terminal 30, information processing may be performed using only the user communication terminal 30.

[0073] In the embodiment of the present invention, an example is shown in which numbers are assigned to characters or label information of text data for training data, but the present invention is not limited to this. For example, the numbers may be alphabets, symbols, or any other identifiable information. [Explanation of symbols]

[0074] 1 Information processing system, 10 Information processing server, 11 Control unit, 12 Memory unit, 12a Text data database, 13 Communication unit, 14 Display unit, 30 User communication terminal, 31 Display unit, 32 Control unit, 33 Memory unit, 34 Operation unit, 35 Communication unit, 100 Information processing control unit, 110 Acquisition unit, 120 Conversion unit, 130 Machine learning unit, 140 Generation unit, 150 Addition unit, 160 Transmission unit, 310 Transmission unit, 330 Reception unit, 500 Text data, 600 Teacher Data, 601...Text data for training data, 610...Column, 611...Text, 620...Column, 621...Label information, 623...Second classification information, 700...Text data, 701...First transformed data table, 702...Text data, 703...Embedding vector data table, 704...Embedding vector, 705...First feature vector, 706...Second feature vector, 707...Estimated value, 708...Second transformed data table, 800...Data set, 810...Text, 820...Label information

Claims

1. The computer Acquire irregular document data in which multiple pieces of character information are arranged in segments, converting the irregular document data into text data in which each of the plurality of pieces of character information is arranged on each line based on a relative positional relationship; generating classification information for classifying each line in the text data into a predetermined segment using a model obtained by machine learning the relationship between character information in the same line, character information arranged in adjacent lines, and segments corresponding to the text data of each line in the text data; The machine learning is supervised machine learning that uses training data obtained by converting training text data into data including first identification information that is identifiable corresponding to the classification information and composed of numbers, and second identification information that is identifiable for an independent character included in the training text data and composed of numbers, to learn relationships between character information in the same line, character information arranged in adjacent lines, and segments corresponding to text data in each line in the text data. Information processing methods.

2. the classification information includes second classification information corresponding to a second segment indicating a grouping of text data in one segment; The information processing method according to claim 1 .

3. generating an embedding vector in the model based on the second identification information using an embedding vector data table configured as a vector corresponding to each character; The information processing method according to claim 2 .

4. generating a first feature vector between characters in the same line and a second feature vector between characters in adjacent lines using the embedding vector in the model; and generating an estimate for estimating the classification information based on the first feature vector and the second feature vector. The information processing method according to claim 3.

5. The irregular document data is resume data, the resume data includes historical information associated with each user; The information processing method according to claim 1 .

6. Acquire irregular document data in which multiple pieces of character information are arranged in segments, converting the irregular document data into text data in which each of the plurality of pieces of character information is arranged on each line based on a relative positional relationship; a control unit that generates classification information for classifying each line in the text data into a predetermined segment by using a model obtained by machine learning of character information in the same line, character information arranged in adjacent lines, and a relationship between segments corresponding to the text data of each line in the text data, The machine learning is supervised machine learning that uses training data obtained by converting training text data into data including first identification information that is identifiable corresponding to the classification information and composed of numbers, and second identification information that is identifiable for an independent character included in the training text data and composed of numbers, to learn relationships between character information in the same line, character information arranged in adjacent lines, and segments corresponding to text data in each line in the text data. Information processing system.

7. the classification information includes second classification information corresponding to a second segment indicating a grouping of text data in one segment; 7. The information processing system according to claim 6.

8. generating an embedding vector in the model based on the second identification information using an embedding vector data table configured as a vector corresponding to each character; The information processing system according to claim 7.

9. generating a first feature vector between characters in the same line and a second feature vector between characters in adjacent lines using the embedding vector in the model; and generating an estimate for estimating the classification information based on the first feature vector and the second feature vector. The information processing system according to claim 8.

10. The irregular document data is resume data, the resume data includes historical information associated with each user; 7. The information processing system according to claim 6.

11. To your computer Acquire irregular document data in which multiple pieces of character information are arranged in segments, converting the irregular document data into text data in which each of the plurality of pieces of character information is arranged on each line based on a relative positional relationship; generating classification information for classifying each line in the text data into a predetermined segment using a model obtained by machine learning the relationship between character information in the same line, character information arranged in adjacent lines, and segments corresponding to the text data of each line in the text data; The machine learning is supervised machine learning that uses training data obtained by converting training text data into data including first identification information that is identifiable corresponding to the classification information and composed of numbers, and second identification information that is identifiable for an independent character included in the training text data and composed of numbers, to learn relationships between character information in the same line, character information arranged in adjacent lines, and segments corresponding to text data in each line in the text data. program.

Images