JP2005165754A - Text mining analysis apparatus, text mining analysis method, and text mining analysis program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform analyses similar to fixed point observation and detection of characteristic tendencies by comparing/editing a plurality of text mining analysis results. <P>SOLUTION: A text mining analysis apparatus comprises more than one text data 21, a feature analysis execution means 31 for applying text mining analyses to the respective text data, and more than one analysis result data 22 for displaying a feature word obtained by the analysis and one or more than one predetermined feature degree corresponding to the feature word. The apparatus has comparison means 32, 33, and 34 which select the analysis result data to be compared and analyzed based on the input information by an analyzer; create a comparison list on the basis of the selected analysis result data; set a comparison condition based on the input information by the analyzer; execute the comparison analysis corresponding to the comparison list according to the comparison condition; and output the comparison result by the comparison analysis. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a text mining technique for extracting useful information by analyzing text data accumulated every day, and in particular, by comparing and editing a plurality of analysis results, a fixed point observation analysis and appearance of a characteristic tendency The present invention relates to a text mining analysis apparatus, a text mining analysis method, and a text mining analysis program.

Conventionally, data including text that is digitized and accumulated every day (hereinafter simply referred to as text data), for example, inquiry histories sent to contact centers and customer satisfaction survey questionnaires, etc. as raw voices from customers, It is used as feedback in the activities of companies.
Such text data is generally organized at daily and monthly timings, and is used for various analyses, and is attracting attention as being useful for marketing, product development, and sales activities in companies. ing.
For this reason, various text mining analysis techniques have been proposed as techniques for extracting useful information based on such text data.

For example, text mining analysis technology that can automatically collect and analyze questionnaire response sentences including free answer descriptions in natural language, and distribute the analysis results to the requester as rule format knowledge Has been proposed (see, for example, Patent Document 1).
In addition, a text mining analysis technique capable of clarifying the tendency of text data accumulated every day and finding characteristic individual data has been proposed (for example, see Patent Document 2).
In addition, a text mining analysis technique that can easily analyze the tendency of a document set by providing a text mining function that can analyze the contents of a document set from a plurality of viewpoints has been proposed ( For example, see Patent Document 3.)

Japanese Laid-Open Patent Publication No. 2001-266060 JP 2002-215647 A JP 2001-318939 A

However, according to these conventional text mining techniques, although feature information can be extracted and classified in each analysis, there is a problem in that these analysis results cannot be compared. For example, there was a problem that it was necessary to manually associate each analysis result with regard to the awareness of the topic that first appeared this month.
In addition, since the analyzed results cannot be reused at different timings, if you want to perform an analysis that includes this analyzed target data, prepare the analytical data that also includes this target data, and then perform text mining again. There is a problem that it is necessary to carry out the analysis and it takes a lot of time for the analysis.

  The present invention is considered in view of the above circumstances, based on a plurality of text mining analysis results, extracting differences and feature points between the analysis results, and outputting a comparison result for displaying these, It is an object of the present invention to provide a text mining analysis apparatus, a text mining analysis method, and a text mining analysis program that can detect fixed point observation of text data and the appearance of a characteristic tendency.

  In order to achieve the above object, a text mining analyzer of the present invention includes two or more text data, a feature analysis execution means for text mining analysis of each text data, a feature word obtained by this analysis, and a feature word A text mining analyzer that includes two or more analysis result data having one or more corresponding predetermined features, and selects and selects analysis result data to be subjected to comparative analysis based on input information A comparison list is created based on the analysis result data, a comparison condition is set based on the input information, a comparison analysis is performed on the comparison list according to the comparison condition, and a comparison result is output according to the comparison analysis. .

If the text mining analyzer is configured in this way, not only text mining analysis can be performed on the current text data, but also the analysis results over the past can be used to perform comparative analysis between the analysis results. Since it can be performed, it becomes possible to efficiently recognize and discover risks and trends.
Enterprises generally analyze customer feedback and use it in business activities, but the amount of text data gathered at a contact center that collects information from customers is often enormous.

For example, when trying to perform a 10-year comparison for the same month, according to the conventional text mining analysis technology, it is necessary to re-execute the text mining analysis for 10 years of data. The resources that need to be used were enormous.
According to the present invention, it is possible to perform text mining analysis using analysis results performed in the past, and it is possible to realize effective text mining analysis very quickly compared to conventional analysis techniques. It becomes.
The predetermined feature level means, for example, the appearance frequency of various feature words, various statistical scales, and the like.

  In the text mining analyzer of the present invention, the comparison means selects one of the analysis result data in the comparison list as a target based on the input information in setting the comparison condition, and sets the comparison target for this target. Specify one or more analysis result data from the comparison list, and detect the feature words that are present in the target but not in the analysis result data to be compared as new words in the comparison analysis. In the output of the comparison result, the comparison result for highlighting the new word is output.

If the text mining analyzer is configured in this way, for example, an analysis result created based on text data input as the latest customer voice can be selected as a target, and the analysis result can be obtained in a predetermined period in the past. Compared with the analysis results, it is possible to grasp newly appearing feature words and efficiently grasp the latest trends and the like.
In addition, because the results obtained as such comparison features are highlighted, that is, output and visualized in a format that is easy to grasp intuitively, not only analysts, but also all on-site personnel in the company, for example Therefore, the comparison result can be easily fed back, and the use of the comparison result can be promoted.

  In the text mining analysis apparatus of the present invention, the comparison unit selects one or more feature words to be analyzed from the comparison list based on the input information in setting the comparison condition, and performs comparison analysis. In the execution, all the feature words to be analyzed in the comparison list are detected, and in the output of the comparison result, the comparison result for highlighting the detected feature words to be analyzed is output.

  If the text mining analyzer is configured in this way, the analyst can easily grasp the situation in which the keyword of interest appears in the current or past analysis results. It is possible to efficiently capture changes in the appearance status.

In the text mining analysis apparatus of the present invention, the comparison unit outputs a graph that displays, for each analysis result data in the comparison list, the feature degree corresponding to the detected feature word to be analyzed in the output of the comparison result. It is as composition to do.
If the text mining analyzer is configured in this way, it is possible to clearly indicate in what situation the keyword of interest has appeared in the current or past analysis results, so that Thus, it is possible to provide a comparison result that can be effectively used for grasping trends and the like.

Further, the text mining analyzer of the present invention sets a predetermined alarm output criterion based on input information, detects a feature word satisfying the alarm output criterion from a comparison list or a comparison result, and generates an alarm for the feature word. An alarm output means for outputting information is provided.
If the text mining analyzer is configured in this way, the analyst can output an alarm on the comparison results based on the various indicators that he or she wants to grasp, so that the comparison results can be clearly recognized. It becomes.
For example, if a feature word with an increase rate of appearance frequency larger than a predetermined value appears, the analyst can set trends and risks more effectively by setting an alarm to be output. It becomes possible to grasp.

In the text mining analyzer of the present invention, the comparison means or the alarm output means is connected to the output device provided in the text mining analyzer or the text mining analyzer by wire or wirelessly the comparison result or alarm information. The information is output to the information processing apparatus.
If the text mining analyzer is configured in this way, the analyst can check the comparison result on the display of the text mining analyzer or print it. Further, by sending the comparison result to an information sharing system such as an enterprise division EIP (Enterprise Information Portal) via a communication line, alarm information as notice information input from the text mining analyzer is quickly reported. It can be used as an alarm.

In the text mining analysis apparatus of the present invention, when the comparison unit receives the selection input of the feature word in the comparison list, the text mining analysis apparatus displays the information of the text data from which the selected feature word is extracted.
When the text mining analyzer is configured in this way, the analyst refers to what text data the feature word in the analysis result displayed in the comparison list or comparison result is described in. The customer's voice can be grasped based on a new word or a keyword of interest.

  The text mining analysis method of the present invention includes two or more text data, a feature analysis execution means for text mining analysis of each text data, a feature word obtained by this analysis, and one or two corresponding to the feature word. A text mining analysis method for performing a comparative analysis of analysis result data using a text mining analysis device having two or more analysis result data possessing the above-mentioned predetermined features. Select analysis result data to be subjected to comparative analysis based on information, create a comparison list based on the selected analysis result data, set comparison conditions based on input information, and execute comparison analysis on comparison list according to comparison conditions This is a method for outputting a comparison result by comparative analysis.

  Further, according to the text mining analysis method of the present invention, the text mining analyzer sets a predetermined alarm output criterion based on the input information, detects a feature word satisfying the alarm output criterion from the comparison list, and detects the feature word. As a method of outputting alarm information about

If the text mining analysis method is used in this way, the analysis results obtained over the past can be used to perform comparative analysis between the analysis results, so that awareness and discovery of risks and trends can be efficiently performed. Can be done.
In other words, by using past analysis results in this way, when performing analysis including past text data, there is no need to perform text mining analysis again using these text data. Resources can be reduced, and extremely quick analysis can be realized.
Furthermore, an alarm can be output when the result of the comparative analysis satisfies a predetermined standard, and the analyst can more effectively grasp risks and trends.

  The text mining analysis program of the present invention includes two or more text data, a feature analysis execution means for text mining analysis of each text data, a feature word obtained by this analysis, and one or two corresponding to the feature word. A text mining analysis program that causes a text mining analysis device having two or more analysis result data possessing the above predetermined features to perform a comparative analysis of the analysis result data, the input information to the text mining analysis device Based on the selected analysis result data, select the analysis result data, create a comparison list based on the selected analysis result data, set the comparison condition based on the input information, and execute the comparison analysis for the comparison list according to the comparison condition And a comparison result by comparison analysis is output.

  The text mining analysis program of the present invention causes the text mining analyzer to set a predetermined alarm output criterion based on input information, detect a feature word that satisfies the alarm output criterion from a comparison list, and Is configured to output alarm information about the.

  If the text mining analysis program is configured in this way, the text mining analyzer can be used to perform comparison analysis between analysis results using past analysis results and output the comparison results. Since an alarm can be output, the analyst can efficiently grasp risks and trends.

According to the present invention, it is possible to perform a text mining analysis on the current text data, and to perform a comparative analysis between the analysis results using the analysis result and the result of the text mining analysis performed in the past. Therefore, it is possible to efficiently recognize and discover risks and trends.
In addition, by this comparative analysis, it is possible to grasp the state of change of newly appearing feature words, noticed keywords, etc., and it is possible to efficiently grasp the latest trends and the like.
Furthermore, the results obtained as such comparison features are output and visualized in an intuitively easy-to-understand format, so that not only the analyst but also all on-site personnel in the company can compare the results. Feedback can be performed easily, and the use of the comparison result can be promoted.
In addition, alarms can be output for comparison results based on various indicators that you want to understand, so you can clearly recognize such comparison results, and analysts can understand trends and risks more effectively. It becomes possible to do.

Hereinafter, a preferred embodiment of a text mining analyzer according to the present invention will be described with reference to the drawings.
Note that the text mining analyzer of the present invention shown in the following embodiments is operated by a computer controlled by a program. The program sends a command to each component of the computer to perform predetermined processing necessary for the operation of the text mining analyzer, such as feature analysis processing, comparison setting processing, comparison list display processing, comparison feature extraction processing, etc. . Thus, each process and operation in the text mining analyzer of the present invention can be realized by specific means in which the program and the computer cooperate.
The program is stored in advance in a recording medium such as a ROM and a RAM, and is executed by causing the computer to read the program from a recording medium mounted on the computer. For example, the program may be read by the computer via a communication line.
Further, the recording medium for storing the program can be constituted by, for example, a semiconductor memory, a magnetic disk, an optical disk, or any other recording means readable by any computer.

[First embodiment]
First, the structure of 1st embodiment of this invention is demonstrated with reference to FIGS. FIG. 1 is a block diagram showing the configuration of the text mining analyzer of this embodiment. 2 to 8 show, in order, text data, analysis result data, comparison target list setting screen, comparison result list, new word extraction display result, attention keyword highlight display result, graph display in the text mining analyzer of this embodiment. It is a figure which shows a result.

As shown in FIG. 1, the text mining analyzer of this embodiment includes an input device 10, a storage device 20, a data analysis processing device 30, and an output device 40.
The input device 10 is a device such as a keyboard used by a person who performs mining analysis on text data (hereinafter simply referred to as an analyst) to input comparative analysis conditions and the like into the data analysis processing device 30. is there.

  The storage device 20 is a device for recording information. As shown in FIG. 1, the storage device 20 has a plurality of text data 21 (text data 21-1,..., Text data 21-n) and a plurality of analysis result data 22. (Analysis result data 22-1,..., Analysis result data 22-m) are accumulated. Note that the text data 21 and the analysis result data 22 do not necessarily correspond one-to-one. If one analysis result data 22 is created based on a plurality of text data 21, The case where a plurality of analysis result data 22 is created is also included.

The text data 21 is negative type data including texts such as answers to questionnaires, inquiries, and complaints, and can be, for example, as shown in FIG.
As shown in the figure, the text data 21 can be a comma-delimited text file called CSV, a file having a certain format extracted from a database such as a relational database, etc., and is composed of columns and rows. Has been.

The text data 21 shown in FIG. 2 is an example of data obtained as a result of conducting a questionnaire for a certain product. The data structure includes “date and time”, “product name”, “gender”, “age category”, “ It has column items such as "evaluation point", "opinion", and "request".
“Date and time” is the date and time information when the data was input, “Product name” is the name of the product that is the subject of the questionnaire, and “Gender” and “Age category” are the attribute information of the survey respondents . The “evaluation score” is the score of the respondent's evaluation for the product (in this example, a five-point scale of 5 points), and the “opinion” and “request” are the opinions and requests of the respondent to the product Is described).

  As mentioned above, it goes without saying that information other than a questionnaire can be used as the text data 21, for example, a database of inquiries and complaints sent to the contact center, and daily reports written by sales representatives. It is also preferable to use a modified one. Further, the number of items in the column of the text data 21 is not limited to that illustrated in FIG. 2 and may be several hundred other items, and the number of lines of the text data 21 is several hundred to several hundreds. It can be as huge as ten thousand.

The analysis result data (hereinafter simply referred to as analysis result) 22 is an execution result output by the feature analysis execution means 31 and can be, for example, as shown in FIG. As shown in the figure, the data structure may have column items such as “characteristic word”, “frequency”, “total frequency”, and “statistical scale”.
It is also preferable to display the analysis result 22 as a diagram or a list, and the corresponding original text (text data 21 from which feature words are extracted) can be referred to from the displayed feature words. Is also preferable. As the method, for example, individual feature words in the analysis result 22 can be selected by, for example, clicking, and when the selection is made, the feature analysis execution means 31 uses the selected feature words as a key for the text data 21. And a window displaying a list of the names of the text data 21 having the characteristic words can be output.

The data analysis processing device 30 is an information processing device that operates under program control, and includes a feature analysis execution unit 31, a comparison setting unit 32, a comparison list display unit 33, and a comparison feature extraction unit 34.
The feature analysis execution means 31 performs text mining analysis on the input text data 21 according to the conditions set from the input device, and outputs an analysis result 22. This feature analysis execution means 31 corresponds to the text classification engine described in Patent Document 1 which is a patent document by the present applicant.
The feature analysis execution unit 31 extracts words from the text data 21, analyzes the dependency, extracts a classification rule from the specific keyword and category, and outputs the result to the analysis result 22 and the output device 40.

For example, FIG. 3 shows the characteristics used in the “opinion” in the data corresponding to the condition of “sex” = “male” and “age category” = “40s” with respect to the text data 21 shown in FIG. The example which displays the result of extracting a typical word is shown.
In the figure, characteristic words such as “friendly” and “young → prefer” are displayed. As a rule for determining a characteristic word, for example, a word with high output frequency is determined to be characteristic. Can do. In addition, in the case of this example, data other than the target category, in the case of “gender” = “female” or “age category” = “other than the 40s”, statistical processing is performed on the characteristic degree, It is also possible to grasp typical words.

  The comparison setting unit 32 sets an analysis target range and a comparison option for performing a comparative analysis of the difference based on the plurality of analysis results 22 output by the feature analysis execution unit 31. The comparison setting means 32 displays a comparison target list setting screen as shown in FIG. 4, and sets the analysis result 22 to be displayed in the comparison list displayed by the comparison list display means 33 based on the selection input information by the analyst. To do. The comparison setting means 32 can also set the display order of the analysis results 22 to be displayed on the comparison list. Although not shown, the comparison setting means 32 can display a setting screen for setting various comparison options. The comparison setting means 34 sets comparison options based on the input information of the analyst. Perform comparative analysis based on options.

The comparison list display means 33 outputs a comparison list according to the conditions set by the comparison setting means 32. FIG. 5 shows an example of a comparison result list that is a comparison list displayed by the comparison list display means 33.
With this user interface, each analysis result can be displayed as a list, and sorting (rearrangement) can be performed on the basis of a frequency or a statistical scale serving as a reference for the characteristic degree. Further, the original text in which the word is used from the feature word, that is, the text data 21 from which the word is extracted can be referred to. The reference method can be the same as in the analysis result 22. The feature level means an index for grasping the feature of the analysis result.

The comparison feature extraction unit 34 extracts differences between the analysis results and feature points from the comparison list created by the comparison list display unit 33 based on a predetermined condition, and outputs the difference as a comparison result. At this time, the comparison feature extraction means 34 can provide a plurality of analysis methods by setting comparison options. In addition, for each comparison option, a comparison condition is set based on information input by the analyst, and a comparison analysis is performed on the comparison list according to the comparison condition.
For example, new words can be extracted from the comparison list and output in the form of a table, a figure, or a list. FIG. 6 shows an example of such a new word extraction display result. In FIG. 6, the feature words that are not displayed in the feature analysis results 1 and 2 but appear for the first time in the feature analysis result 3 are shown. Is highlighted to be distinguishable from other words.

  It is also possible to highlight the attention word and graph it. FIG. 7 shows an example of a highlighted display result of a keyword of interest. In FIG. 7, “design” is highlighted so as to be distinguishable from other words. Of course, a plurality of feature words can be highlighted. FIG. 8 shows a display result of a graph that displays the change in the feature level of the keyword of interest selected in this way. In the example of the figure, the frequency of appearance of the four keywords “design”, “trust”, “ordinary”, and “difficult to use” in September to December, which is the month corresponding to the feature analysis results 1 to 4, is shown. The change is displayed.

  The output device 40 is a device such as a display device or a printing device that outputs the analysis result 22, the comparison list created by the comparison list display means 33, and various tables and graphs created by the comparison feature extraction means 34.

Next, a processing procedure in the text mining analyzer of the present embodiment will be described with reference to FIG. 1 and FIGS.
FIG. 9 to FIG. 11 are flowcharts showing the comparison setting processing procedure, the new word extraction processing procedure, and the attention keyword highlight / graphing processing procedure in the text mining analyzer of this embodiment in order.

First, the analyst designates the text data 21 to be analyzed from the input device 10, sets the analysis conditions, executes the analysis by the feature analysis execution means 31, and creates the analysis result 22. Is displayed on the output device 40.
At this time, the feature analysis execution means 31 cuts out words from the text data 21, analyzes the dependency, and classifies based on the specific keyword and category based on the specification of the text data 21 input from the input device 10 and the analysis conditions. The rule is extracted, and an analysis result 22 having the result is created in the storage device 20 and output to the output device 40 for display.

In general, with regard to the history of inquiries sent to the contact center and customer satisfaction survey questionnaires, etc., feature analysis is usually performed on a daily or monthly basis for the purpose of providing feedback to company activities. Are sequentially stored and accumulated in the storage device 20.
Then, based on the analysis results accumulated in this way and the analysis results 22 for the text data 21 that is the object for which the current feature is to be grasped, a fixed point observation analysis, the appearance of a characteristic trend, etc. When performing, the comparison setting means 32 is used to set conditions for displaying a comparison list.

  At this time, as shown in FIG. 9, when the analyst selects an analysis result to be subjected to comparative analysis from the accumulated analysis results 22, the comparison setting means 32 inputs this selection information (step A1). 4 is added to the comparison target list as shown in FIG. 4 (step A2). These operations are repeated until selection of a desired analysis result is completed (step A3).

Furthermore, when the analyst selects one analysis result from the comparison target list for which selection has been completed, and designates and sets the order of comparison, the comparison setting means 32 inputs the selection information (step A4). The order setting is executed (step A5). These operations are repeated until the setting is completed for all the analysis results 22 in the comparison target list (step A6).
This setting of the comparison order is particularly effective when it is desired to make sense in the order of comparison, such as when trend analysis is performed in chronological order or when analysis is performed based on a customer loyalty cycle.

Next, the comparison list display means 33 displays a comparison list as shown in FIG. 5 based on the conditions set as a comparison target list by the comparison setting means 32.
Further, the comparison feature extraction means 34 displays various comparison results as shown in FIGS. 6 to 8 on the output device 40 for the differences between the analysis results and the feature points based on the options set by the analyst. .

FIG. 10 is a flowchart showing a processing procedure for extracting a new word, which is an option of processing by the comparison feature extraction means 34.
When an analyst selects a target analysis result (hereinafter simply referred to as a target result) from the analysis results displayed in the comparison list as a comparison condition, and specifies a new word extraction range, The comparison feature extraction unit 34 inputs selection information of the target result selected by the analyst (step B1), and inputs designation information of a new word extraction range designated by the analyst (step B2). This target result means an analysis result as a target for determining whether or not a new word appears. The new word extraction range means an analysis result as an object for determining whether or not a feature word in the target result appears.

Next, the comparison feature extraction unit 34 selects one word output as a feature word in the target result (step B3), and the selected feature word is the analysis result specified in the new word extraction range. Is not output as a feature word, and if it does not appear, it is determined that the selected feature word is a new word (step B4).
Then, display switching information for the selected feature word is created, and a new word extraction display result obtained by converting the display of the selected feature word into a highlighted display that can be distinguished from other feature words in the comparison list. Create (step B5). In addition, you may create the list, figure, etc. which display only this new word.
The above operation is repeated for all feature words in the target result (step B6), and the completed word extraction display result is output.

  The word extraction display result shown in FIG. 6 shows an example in which feature analysis result 3 is selected as the target result and feature analysis results 1 and 2 are designated as the new word extraction range. In other words, in this case, the processing setting is such that only feature words that do not appear in the feature analysis results 1 and 2 but appear for the first time in the feature analysis result 3 are extracted. As shown in FIG. Words such as “Store” and “Salesperson” are highlighted as new words.

FIG. 11 is a flowchart showing a processing procedure for highlight-outputting a keyword of interest, which is another option of processing by the comparison feature extraction unit 34.
When an analyst selects, as a comparison condition, a word of interest from the analysis results displayed in the comparison list as an attention keyword that is a word to be analyzed, and adds it to a highlight target list (not shown) The comparison feature extraction means 34 inputs this attention keyword selection information (step C1) and adds it to the highlight target list (step C2). This is repeated for the number of desired keywords of interest (step C3).

Then, the comparison feature extraction unit 34 searches for whether or not each keyword added to the highlight target list is displayed as a feature word in the analysis result in the comparison list. The display switching information about is generated, and the display is converted into a highlight display and displayed (step C4).
Further, the comparison feature extraction unit 34 creates a graph that displays the change in the feature level for each keyword in the highlight target list based on the search result, and displays the graph on the output device 40 (step C5).

  The attention keyword highlight display result shown in FIG. 7 shows an example in which “design” is added to the highlight target list, and the position of “design” is highlighted on each analysis result in the comparison list. Similarly, FIG. 8 shows an example in which “reliable”, “ordinary” and “difficult to use” are added to the highlight target list and graphed. In FIG. 8, a line graph is displayed in which the transition of the keyword characteristic degree can be listed. Of course, the type of the graph is not limited to the line graph, and it is also preferable that the graph can be appropriately changed to various other graphs.

[Second Embodiment]
Next, a second embodiment of the present invention will be described with reference to FIG. FIG. 2 is a block diagram showing the configuration of the text mining analyzer of this embodiment.
The present embodiment is different from the first embodiment in that alarm information is output when a feature appears exceeding a predetermined alarm output standard set by an analyst.
The text mining analyzer of this embodiment has an alarm output means 35 in the data analysis processor 30 in addition to the configuration of the first embodiment shown in FIG. In addition, an information sharing server 50 capable of receiving output information from the alarm output means 35 by wire or wirelessly is provided.

The alarm output means 35 sets a predetermined alarm output standard based on information input by an analyst, detects a word satisfying this alarm output standard from the comparison list, and outputs the alarm information.
In the present embodiment, the alarm output means 35 extracts a feature word having a characteristic degree exceeding the threshold information from each analysis result in the comparison list based on the predetermined threshold information input by the analyst, and generates an alarm. Information is created and output to the output device 40 and the information sharing server 50.

The threshold information can be various condition information. For example, it can be the amount of change of the feature degree, and can be the increase rate or decrease rate of the appearance frequency. Also, the absolute value of the feature degree can be used, and a feature word exceeding a predetermined total appearance frequency can be set as an alarm output target. Further, whether or not a specific keyword set in advance appears can be set as an alarm output target. These can be set as options in the alarm output means 35.
The alarm information is not particularly limited as long as it can give a warning to the analyst. For example, a list of feature words extracted as alarm output targets in a warning window is displayed. It can be.

  The information sharing server 50 may be an information processing apparatus that can utilize alarm information as notice information input from a text mining analyzer as a breaking alarm in cooperation with an information sharing system such as a corporate sector EIP. it can.

Next, an alarm output processing procedure in the text mining analyzer of this embodiment will be described with reference to FIG. FIG. 6 is a flowchart showing an alarm output processing procedure in the text mining analyzer of the present embodiment, and shows an alarm output processing procedure for a keyword appearance increase rate that is one option of processing by the alarm output means 35.
First, the analyst selects a target result from the comparison list displayed by the comparison list display means 33, the new word extraction display result displayed by the comparison feature extraction means 34, the attention keyword highlight display result, etc. Is set, the alarm output means 35 inputs selection information of the selected target result (step D1) and sets a threshold (step D2).

Next, the alarm output means 35 selects one feature word from the selected target result based on the input information by the analyst (step D3), the appearance frequency of the feature word in the target result, and the previous frequency. It compares with the appearance frequency of the feature word in an analysis result, and determines whether the appearance increase rate of the feature word exceeds the set threshold value (step D4).
If the threshold is exceeded, the alarm output means 35 generates alarm information for the feature word (step D5).
Then, for all feature words included in the target result, the process of determining whether the appearance frequency exceeds the threshold is repeated (step D6), and if alarm information is generated, these alarm information Is output to the output device 40 and the information sharing server 50 (step D7).

The previous analysis results used in the comparison between the appearance frequency of feature words in the target results and the appearance frequency of feature words in the previous analysis results are those before the target results existing in the comparison list that is the target of the comparative analysis. All analysis results of can be used.
Also, for example, in step D1, as in the first embodiment, together with the selection of the target result, it is possible to specify one or more analysis results that are symmetrical for comparison, and the appearance frequency of feature words in the target result It is also preferable to perform comparison with the appearance frequency of feature words in the designated analysis result.

In addition, this invention is not limited to the above embodiment, It cannot be overemphasized that a various change implementation is possible within the scope of the present invention.
For example, in the above embodiment, the comparison feature extraction means 34 is configured to perform the new word extraction display, the attention keyword highlight display, and the graph display on the output device 40. However, based on more comparative analysis indices. Of course, the result may be displayed. It is also preferable that these comparative analysis results be transmitted to the information sharing server 50 so that they can be output.
Further, the alarm output means 35 can transmit the alarm information and the comparison list can be transmitted to the information sharing server 50. Other modifications can be made as appropriate.

  According to the present invention, there is provided an information mining device or an information mining device for extracting various information such as a trend or tendency to be noticed, a risk or a problem from a customer's voice such as an inquiry sent to a contact center or a customer satisfaction survey. The present invention can be applied to uses such as a program for realizing on a computer.

It is a block diagram which shows the structure of the text mining analyzer of 1st embodiment of this invention. It is a figure which shows the text data in the text mining analyzer of 1st embodiment of this invention. It is a figure which shows the analysis result data in the text mining analyzer of 1st embodiment of this invention. It is a figure which shows the comparison object list setting screen in the text mining analyzer of 1st embodiment of this invention. It is a figure which shows the comparison result list in the text mining analyzer of 1st embodiment of this invention. It is a figure which shows the new word extraction display result in the text mining analyzer of 1st embodiment of this invention. It is a figure which shows the attention keyword highlight display result in the text mining analyzer of 1st embodiment of this invention. It is a figure which shows the graph display result in the text mining analyzer of 1st embodiment of this invention. It is a flowchart which shows the comparison setting process sequence in the text mining analyzer of 1st embodiment of this invention. It is a flowchart which shows the new word extraction process procedure in the text mining analyzer of 1st embodiment of this invention. It is a flowchart which shows the attention keyword highlight graphing process sequence in the text mining analyzer of 1st embodiment of this invention. It is a block diagram which shows the structure of the text mining analyzer of 2nd embodiment of this invention. It is a flowchart which shows the alarm output process sequence in the text mining analyzer of 2nd embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Input device 20 Storage device 21 (21-1 to 21-n) Text data 22 (22-1 to 22-m) Analysis result data 30 Data analysis processing device 31 Feature analysis execution means 32 Comparison setting means 33 Comparison list display means 34 comparison feature extraction means 35 alarm output means 40 output device 50 information sharing server

Claims (11)

Two or more text data, feature analysis execution means for text mining analysis of each text data, and two or more possessing feature words obtained by this analysis and one or more predetermined feature degrees corresponding to the feature words A text mining analyzer with analysis result data of
Select the analysis result data to be subjected to the comparative analysis based on the input information, create a comparison list based on the selected analysis result data, set the comparison condition based on the input information, and perform the comparison list according to the comparison condition. A text mining analysis apparatus comprising: a comparison unit that executes comparison analysis and outputs a comparison result of the comparison analysis. The comparing means is
In the setting of the comparison condition, one of the analysis result data in the comparison list is selected as a target based on input information, and one or more analysis result data to be compared with respect to the target is selected in the comparison list. Specify from
In the execution of the comparative analysis, a feature word that exists in the target but does not exist in the analysis result data to be compared is detected as a new word,
The text mining analysis apparatus according to claim 1, wherein in the output of the comparison result, a comparison result for highlighting the new word is output. The comparing means is
In setting the comparison condition, based on the input information, select one or more feature words to be analyzed from the comparison list,
In the execution of the comparative analysis, all feature words to be analyzed in the comparison list are detected,
The text mining analysis apparatus according to claim 1 or 2, wherein in the output of the comparison result, a comparison result for highlighting the detected feature word to be analyzed is output. In the output of the comparison result, the comparison means
The text mining analysis apparatus according to claim 3, wherein a graph displaying the feature degree corresponding to the detected feature word to be analyzed for each analysis result data in the comparison list is output. The text mining analyzer according to any one of claims 1 to 4,
Alarm output means for setting a predetermined alarm output standard based on input information, detecting a feature word satisfying the alarm output standard from the comparison list or the comparison result, and outputting alarm information for the feature word A text mining analyzer characterized by having. The comparison means or the alarm output means outputs the comparison result or the alarm information to an output device provided in the text mining analyzer or an information processing device connected to the text mining analyzer by wire or wirelessly. The text mining analyzer according to any one of claims 1 to 5, wherein: The comparing means is
The text mining according to any one of claims 1 to 6, wherein when the selection input of the feature word in the comparison list is received, the information of the text data from which the selected feature word is extracted is displayed. Analysis equipment. Two or more text data, feature analysis execution means for text mining analysis of each text data, two or more possessing feature words obtained by this analysis and one or more predetermined feature degrees corresponding to the feature words A text mining analysis method for performing a comparative analysis of the analysis result data using a text mining analysis device provided with the analysis result data of
The text mining analyzer is
Select the analysis result data to be subjected to comparative analysis based on the input information,
Create a comparison list based on the selected analysis result data,
Set comparison conditions based on the input information,
Performing a comparative analysis on the comparison list according to the comparison conditions;
A text mining analysis method, comprising: outputting a comparison result by the comparison analysis. The text mining analyzer is
Based on the input information, set the predetermined alarm output criteria,
Detect feature words that meet the alarm output criteria from the comparison list,
The text mining analysis method according to claim 8, wherein alarm information about the feature word is output. Two or more text data, feature analysis execution means for text mining analysis of each text data, and two or more possessing feature words obtained by this analysis and one or more predetermined feature degrees corresponding to the feature words A text mining analysis program for causing a text mining analysis device comprising the analysis result data to perform a comparative analysis of the analysis result data,
In the text mining analyzer,
Based on the input information, the analysis result data to be subjected to comparative analysis is selected,
Create a comparison list based on the selected analysis result data,
Let the comparison conditions be set based on the input information,
Performing a comparative analysis on the comparison list according to the comparison conditions;
A text mining analysis program for causing a comparison result by the comparison analysis to be output. In the text mining analyzer,
Based on the input information, set a predetermined alarm output standard,
Let the feature words that meet the alarm output criteria be detected from the comparison list,
The text mining analysis program according to claim 10, wherein the alarm information for the feature word is output.

Images