KR102240303B1 - Method and electronic device for producing natural language from data table - Google Patents

Abstract

The present invention relates to a method for generating a natural language from a data table, and an electronic device for generating a natural language from a data table. According to an embodiment of the present invention, the method comprises the steps of: receiving, by a processor, setting information to generate a natural language for analyzing a data table comprising a plurality of fields; checking whether at least one or more natural language generating fields according to settings exist in a read data table; generating and feeding back at least one virtual field which is a natural language generating field which is not included in a data table read for natural language generation when a natural language generating field does not exist; and generating a natural language which analyzes the data table according to the settings in the presence of the natural language generation field. In the case of virtual field feedback, in the step of checking whether the natural language generation field exists, it is checked whether the natural language generation field exists in the data table including the virtual field. The step for generating natural language includes the steps of: generating an entire connecting word applied to all records in the natural language generation field according to a setting; and getting at least one of a field value and a field name of the natural language generation field according to the setting, wherein the field includes a field name and at least one or more field values. Also, provided is an electronic device for generating a natural language from a data table according to the method for generating a natural language from a data table. Therefore, the present invention is to eliminate the cost and time inefficiency that occurs in a collaborative system between a programmer and a data analysis expert.

Description

A method for generating natural language from a data table and an electronic device that performs natural language generation from a data table TECHNICAL FIELD [METHOD AND ELECTRONIC DEVICE FOR PRODUCING NATURAL LANGUAGE FROM DATA TABLE]

The present invention relates to a natural language generation method and an electronic device that performs the same, among others, to a natural language generation method from a data table, and an electronic device that performs natural language generation from a data table according to the method. Specifically, the present invention relates to a method for easily generating a semantic interpretation rule in natural language when a person interprets the meaning of data included in a data table, and an electronic device that performs the same. The present invention relates to a method for generating natural language from a data table and an electronic device that performs the method for supporting the data analysis expert to easily generate natural language results after analyzing data.

The technical field of the present invention belongs to natural language processing technology in the field of artificial intelligence. Natural language processing technology includes Opinion Mining, which analyzes user's evaluation information for products in written text, and Personal Assistant such as Apple's Siri or IBM, which improves usability by providing a natural language interface. Question Answering technology that provides the correct answer to a given question, such as Watson of the company, Curation technology that analyzes the content of articles that users enjoy in the service and places their preferred content first, etc. In addition, there is also a natural language generation technology that expresses data such as numerical values and statistics in natural text so that people can easily understand them. At this time, the present invention corresponds to a natural language generation technology among natural language processing technologies.

Natural language refers to the language used by humans to communicate with each other in daily life.

Conventionally, in the field of information and communication, in order to generate the meaning of data in a natural language, it is used, for example, by hard-coding it into a program. For example, in the case of “what is monthly income”, it is expressed as “monthly income is ()” in the program source, and the value corresponding to the “( )” value is read from the data and displayed. This method is a method developed by a programmer so that an expert who interprets the meaning of data writes a phrase and a phrase written by an expert can be written, that is, a method performed by a data analysis expert and a programmer through collaboration.

In such a conventional method, the difference in terminology and data interpretation method for each task are different depending on the applied task. Therefore, in order to generate natural language, a data analysis expert and a programmer must always develop a new one through collaboration, which takes a lot of time and cost. have.

In addition, conventionally, when the same content is developed in Korean, it takes a lot of time and cost because it has to be re-developed from the beginning in order to apply it in English or Japanese.

Republic of Korea Patent Publication No. 10-2018-0077689 (published on July 9, 2018) Korean Patent Application Publication No. 10-2019-0019748 (published on February 27, 2019) Republic of Korea Patent Publication No. 10-2019-0011031 (published on February 1, 2019)

In order to overcome the shortcomings of the conventional natural language generation method, in order to eliminate the cost and time inefficiency incurred when the programmer and data analysis expert collaborate, only a data expert who is not proficient in programming can generate natural language. In addition, there is a need for a method that can easily and universally generate natural language regardless of the type of work applied, and regardless of languages such as Korean, Japanese, and English.

In the conventional natural language generation method, it is intended to solve the problem that needs to be re-implemented from the beginning every time when natural language is generated, and by presenting a general natural language generation method irrespective of the task. In addition, it is intended to solve this problem by a method in which programming for natural language generation is automatically generated by simple manipulations for natural language generation without complicated programming by personnel who are not proficient in programming.

In order to achieve the above object, in accordance with one aspect of the present invention, the method comprising: receiving, in a processor, setting items to generate a natural language for analyzing a data table including a plurality of fields; Checking whether at least one or more natural language generation fields according to setting items exist in the read data table, the natural language generation field being a field to generate natural language; Generating and feeding back at least one virtual field that is a natural language generation field that is not included in a data table read for natural language generation when the natural language generation field does not exist; And generating a natural language for analyzing the data table according to the settings in the processor when the natural language generation field exists, and in the step of checking whether the natural language generation field exists in the case of feedback from the virtual field, data including the virtual field The step of checking whether the natural language generation field exists in the table and generating the natural language may include: generating a global connection word applied to the entire record of the natural language generation field according to the settings; And obtaining at least one or more of a field value and a field name of a natural language generation field according to a setting, wherein the field includes a field name and at least one field value. Is suggested.

In one example, in the step of receiving the setting items, the setting items include the natural language generation criteria for generating natural language, the natural language generation criteria include all related words and the natural language generation field, and the natural language generation field The matters on include at least one of getting a field value from a natural language generation field and getting a field name of a natural language generation field that meets the field value calling condition, and in the step of checking whether a natural language generation field exists, the field to be checked is It may be any one or more of a field to which the field value is fetched or a field that holds a field value included in a field value call condition.

In this case, in one example, each virtual field is a field created when the field to be checked does not exist, and holds field values generated by processing, processing, or evaluating field values of at least one field of the data table, and field value processing Is one of arithmetic processing of numerical field values of two or more fields, standardization of numerical field values in one field, separation degree or rank processing, and field value processing is at least 2 in fields created by processing existing fields or field values. Field values are reclassified using the grouping of the above field values, or averaged or standardized processing for field values of one field value and another numerical field in the same group, and field value evaluation is used to process existing fields and field values. For one or more fields that have at least a numeric field value in one or more of the fields generated by the field generated by and the field generated by field value processing, it is an evaluation based on the range of the numeric field value, and field value processing or field value evaluation is a setting item. In the step of receiving the input, it is performed according to the virtual field creation criterion included in the setting items, and field value processing may be performed according to a preset definition for the generated virtual field name or according to a user's setting.

At this time, in another example, the operation processing of the numerical field value is an operator operation using an operator, and the standardization processing of the numerical field value is the percentile score, Z score, T score, and the converted score converted from the corresponding score in one field. This is the standardization processing used, and the separation processing of numerical field values is the separation processing using either the average value, the maximum value, the minimum value, and the median number as a reference value or the separation rate increase/decrease rate in one field, and the ranking processing of the numerical field value is one The field is processed according to any one of the ascending order, descending order, and order order in the current data, and the fields subject to standardization of numerical field values, separation degree, and ranking processing are fields created according to existing fields or calculation processing, and field values The reclassification processing for reclassification is character processing according to reclassification or numerical processing according to reclassification, and average or standardization processing during field value processing and field value processing, and numerical processing during reclassification processing for field values, respectively, are numerical values virtual fields. And, each of the character processing and field value evaluation during reclassification processing for the field value can create a dimension virtual field.

In addition, in one example, the field value calling condition relates to any one of matching with a specific value of a field, selective matching of two or more specific values, and satisfaction of a specific range of field values, and natural language generation criteria in the step of generating natural language. Generating the natural language according to the processing order included in the and generating the natural language may include a process of changing a line for the next sentence when the generated natural language sentence is terminated.

In another example, when there are a plurality of natural language generating fields and a plurality of generated virtual fields, a process of generating a natural language for the generated virtual field after generating one virtual field may be repeated.

In addition, in an example, when there are a plurality of natural language generation fields and a plurality of virtual fields to be generated, natural language may be generated for a plurality of natural language generation fields including a plurality of virtual fields generated after the generation of the plurality of virtual fields.

Next, in order to achieve the above object, according to another aspect of the present invention, in an electronic device including at least one processor, the processor generates a natural language for analyzing a data table including a plurality of fields. In order to do so, it receives the setting information, checks whether at least one natural language generation field according to the setting item exists in the read data table, and the natural language generation field is a field to generate natural language, and when the natural language generation field does not exist, natural language At least one virtual field, which is a natural language generation field not included in the data table read for generation, is generated and fed back, and in the case of virtual field feedback, it is checked whether a natural language generation field exists in the data table containing the virtual field, and natural language In the presence of the generated field, a natural language that analyzes the data table is generated according to the settings, but the process of creating a global link that is applied to the entire record of the natural language generation field according to the settings and the field value of the natural language generation field according to the settings , An electronic device for generating natural language from a data table, characterized in that a natural language is generated including a process of obtaining at least one of the field names, and the field includes a field name and at least one field value is proposed. .

In one example, the settings received from the processor include natural language generation criteria for generating natural language, and the natural language generation criteria include information on all linked words and natural language generation fields, and details on natural language generation fields. Includes at least one of getting a field value from a natural language generation field and getting a field name of a natural language generation field that meets the condition of calling the field value, and the field to be checked, which is a natural language generation field to be checked for existence in the processor, is a field. It may be one or more of a field that is a target of getting a value or a field that holds a field value included in a field value calling condition.

In this case, in one example, the processor creates a virtual field that holds the field value generated by processing, processing, or evaluating the field value of at least one field of the data table when the field to be checked does not exist, and the field value processing is It is one of arithmetic processing of numerical field values of two or more fields, standardization of numerical field values in one field, separation degree or rank processing, and field value processing is at least two or more in fields created by processing existing fields or field values. Field values are reclassified using grouping of field values, or averaged or standardized for field values of one field value and another numerical field in the same group, and field value evaluation is performed by processing existing fields and field values. Evaluation based on the range of numerical field values for one or more fields that have at least a numerical field value in one or more of the generated fields and fields created by field value processing, and field value processing or field value evaluation are input settings It is performed according to the virtual field generation criterion included in the item, and field value processing may be performed according to a preset definition for the generated virtual field name or according to a user's setting.

At this time, in another example, the operation processing of the numerical field value is an operator operation using an operator, and the standardization processing of the numerical field value is the percentile score, Z score, T score, and the converted score converted from the corresponding score in one field. This is the standardization processing used, and the separation processing of numerical field values is the separation processing using either the average value, the maximum value, the minimum value, and the median number as a reference value or the separation rate increase/decrease rate in one field, and the ranking processing of the numerical field value is one The field is processed according to any one of the ascending order, descending order, and order order in the current data, and the fields subject to standardization of numerical field values, separation degree, and ranking processing are fields created according to existing fields or calculation processing, and field values The reclassification processing for reclassification is character processing according to reclassification or numerical processing according to reclassification, and the processor is numerical processing according to the average or standardization processing during field value processing and field value processing, and numerical processing during reclassification processing for field values. A value virtual field can be created, and a dimension virtual field can be created according to each character processing and field value evaluation during reclassification processing for the field value.

In addition, in one example, the field value calling condition relates to any one of matching with a specific value of the field, selective matching among two or more specific values, and satisfying a specific range of the field value, and the processor is processed to be included in the natural language generation criteria. The natural language is generated according to the ranking, and the processor may perform a process of changing lines for the next sentence when the generated natural language sentence is terminated.

In another example, when there are a plurality of natural language generation fields and a plurality of generated virtual fields, the processor may repeat the process of generating a natural language for the generated virtual field after generating one virtual field. Alternatively, in another example, the processor may generate natural language for a plurality of natural language generation fields including a plurality of virtual fields generated after generation of a plurality of virtual fields, when there are a plurality of natural language generation fields and a plurality of virtual fields to be generated. I can.

According to an example of the present invention, non-computational personnel who are not proficient in programming, such as a data analyst, perform programming for natural language generation automatically with a simple operation for natural language generation without complex programming to minimize time and work. There is an advantage in that minimization of cost can be achieved.

In addition, by solving the problem of re-implementing the conventional natural language creation every time applied from the beginning, it can be easily applied to other tasks by presenting a general natural language generation method regardless of the task, thus minimizing time and cost. There is an advantage.

In addition, even if a natural language is generated for a specific task due to the difference in the order of arrangement of languages such as Korean, English, Chinese, Japanese, etc., in order to apply it to another language, the problem of having to be re-developed from the beginning has been solved. There is an advantage of minimizing time and cost by presenting a general-purpose natural language generation method that can be easily generated.

1 is a flowchart schematically illustrating a method of generating natural language from a data table according to an embodiment of the present invention.
2 is a flowchart schematically illustrating a method of generating natural language from a data table according to another embodiment of the present invention.
3 is a flowchart schematically illustrating a method of generating natural language from a data table according to another embodiment of the present invention.
4A is a diagram schematically illustrating generation of one virtual field in a method for generating natural language from a data table according to an embodiment of the present invention.
4B is a diagram schematically illustrating another virtual field generation in a natural language generation method from a data table according to an embodiment of the present invention.
4C is a diagram schematically illustrating another virtual field generation in a natural language generation method from a data table according to an embodiment of the present invention.
4D is a diagram schematically illustrating another virtual field generation in a natural language generation method from a data table according to an embodiment of the present invention.
4E is a diagram schematically illustrating another virtual field generation in a natural language generation method from a data table according to an embodiment of the present invention.
4F is a diagram schematically illustrating another virtual field generation in a natural language generation method from a data table according to an embodiment of the present invention.
4G is a diagram schematically illustrating another virtual field generation in a natural language generation method from a data table according to an embodiment of the present invention.
5A is a diagram schematically illustrating a natural language generation in a method of generating a natural language from a data table according to an embodiment of the present invention.
5A is a diagram schematically illustrating a natural language generation in a method of generating a natural language from a data table according to an embodiment of the present invention.
5B is a diagram schematically illustrating another natural language generation in a natural language generation method from a data table according to an embodiment of the present invention.
5C is a diagram schematically illustrating another natural language generation in a natural language generation method from a data table according to an embodiment of the present invention.
5D is a diagram schematically illustrating another natural language generation in a natural language generation method from a data table according to an embodiment of the present invention.
5E is a diagram schematically illustrating another natural language generation in a natural language generation method from a data table according to an embodiment of the present invention.
5F is a diagram schematically illustrating another natural language generation in a natural language generation method from a data table according to an embodiment of the present invention.
6 is a schematic diagram illustrating an electronic device for generating natural language from a data table according to an embodiment of the present invention.
7 is a schematic configuration diagram illustrating a processor of an electronic device that generates natural language from a data table according to an embodiment of the present invention.

Embodiments of the present invention for achieving the above object will be described with reference to the accompanying drawings. In the present description, the same reference numerals mean the same configuration, and a secondary description may be omitted in order to promote an understanding of the features of the present invention to those of ordinary skill in the art.

In the present specification, when one component forms a coupling relationship such as connection or coupling in a relationship with another component, or a transmission relationship such as transmission or transfer, unless there is a limitation of'direct', a'direct' coupling relationship Not only in the form of a transmission relationship, but also in the form of a coupling relationship by a medium or a transmission relationship through a transmission relationship, as another component is related therebetween. In addition, the same applies when terms that can imply the meaning of'contact' such as'top','above','bottom', and'below' are included. In addition, terms indicating direction should be interpreted as a relative concept to the element on which they are referenced.

In addition, it should be noted that even though the configuration is expressed in the singular in the present specification, it may be used as a concept representing all of a plurality of configurations unless interpreted contrary to or contradictory to the concept of the invention.

In addition, in this specification, the description of words such as'comprise' and'comprising' and terms derived therefrom may indicate the possibility of addition, combination, or combination of one or more other elements to the original element or elements. It is not excluded, and further, the description of words having meanings such as'to have','to be composed', and terms derived from them are also the addition, combination, or combination of one or more other elements to the original elements or elements. The possibility of addition or combination of one or more other components should not be excluded if the original element or elements are not lost in their features, functions, and/or properties.

In the present invention, in order to overcome the shortcomings of the conventional natural language generation method, in order to eliminate the cost and time inefficiency incurred when a programmer and a data interpretation expert collaborate, a natural language can be generated even by a data expert who is not proficient in programming. Not only does it not only make it possible, but it also makes it possible to create natural language universally, regardless of the type of work applied, regardless of language such as Korean, Japanese, or English. In the present invention, in the conventional natural language generation method, a problem that needs to be re-implemented from the beginning for each application task when natural language is generated, for example, in order to generate a stock market-related natural language, a natural language that reflects stock market-related terminology should be generated. In order to generate natural language related to commercial area analysis, natural language should be generated that reflects terminology related to commercial area analysis, and natural language that reflects technical terms related to business area must be generated to generate natural language related to business area analysis. It solves the problem that the algorithm cannot be applied to other fields, and proposes a method that automatically generates programming for natural language generation with simple manipulation for natural language generation without complex programming, thus providing a technology that enables general natural language generation regardless of work. present. In addition, in the conventional natural language generation method, even if a specific company generates a natural language for a specific task due to the difference in the order of arrangement of languages such as Korean, English, Chinese, and Japanese, it is a problem that needs to be re-developed from the beginning in order to apply it to other languages. However, in the present invention, a general natural language generation method capable of easily generating natural language regardless of language is proposed to solve this problem.

[How to generate natural language]

First, a method of generating natural language from a data table according to an example of the present invention will be described in detail with reference to the drawings. In this case, a description of an electronic device that generates natural language from a data table according to an example of the invention to be described later may be referred to.

1 is a flowchart schematically illustrating a method of generating natural language from a data table according to an embodiment of the present invention, and FIG. 2 is a schematic flowchart illustrating a method of generating natural language from a data table according to another embodiment of the present invention. FIG. 3 is a flowchart schematically illustrating a method of generating natural language from a data table according to another embodiment of the present invention. Each of FIGS. 4A to 4G is a diagram showing a schematic description of the generation of one virtual field in a method for generating natural language from a data table according to an embodiment of the present invention, and each of FIGS. 5A to 5F is an implementation of the present invention. A diagram schematically illustrating the generation of one natural language in a method of generating a natural language from a data table according to an example.

1 to 3, a method of generating a natural language from a data table according to an example includes a processor (refer to reference numeral 10 in FIGS. 6 to 7), for example, an electronic device having at least one processor 10 (FIG. 6). It may be performed in the processor 10 of (see reference numeral 1). In this case, the electronic device 1 may be a PC, notebook, tablet, or portable terminal, or may be a system device including them.

1 to 3, a method of generating a natural language from a data table according to an example includes inputting settings (S100, S100'), checking the existence of a natural language generation field (S200, S200'), and generating a virtual field ( S300, S300') and the natural language generation step (S400, S400', S400"). In the setting information input step (S100, S100'), in the processor 10, data including a plurality of fields In order to generate a natural language that analyzes the table, setting items are input In the step of confirming the existence of the natural language generation field (S200, S200'), at least one or more natural language generation fields according to settings in the read data table in the processor 10 At this time, the natural language generation field is a field that is a target for generating natural language In the virtual field generation step (S300, S300'), in the processor 10, for natural language generation when the natural language generation field does not exist, At least one virtual field, which is a natural language generation field not included in the read data table, is generated and fed back In the natural language generation step (S400, S400'), in the processor 10, when the natural language generation field exists, the processor 10 Generates a natural language that analyzes the data table according to the settings in. At this time, in the step of checking whether the natural language generation field exists in the case of feedback of a virtual field (S200, S200'), the processor 10 includes a virtual field. It is checked whether a natural language generation field exists in the data table to be used, in which case, the field includes a field name and at least one field value.

In addition, the natural language generation step (S400, S400') includes a whole connection word generation step (S410, S410') and a field value/field name import step (S430, S430'). In the whole connection word generation step (S410, S410'), a global connection word that is applied to the entire record of the natural language generation field is generated according to the settings, and in the field value/field name import step (S430, S430'), the settings are At least one of a field value and a field name of the corresponding natural language generation field is retrieved.

Let's look at each step in detail with reference to the drawings.

Setting information input step (S100, S100')

First, with reference to FIGS. 1 to 3, the setting information input step (S100, S100') will be described. In the setting information input step (S100, S100'), the processor 10 receives the setting information in order to generate a natural language for analyzing a data table including a plurality of fields. In this case, the field includes a field name and at least one field value.

For example, referring to FIG. 2, in the step S100' of receiving the setting item, the setting item includes a natural language generation criterion for generating a natural language. In this case, the natural language generation criterion may include matters about the entire linked language and matters about the natural language generation field. The matters regarding the natural language generation field may include at least one of fetching a field value from the natural language generation field and fetching a field name of a natural language generation field meeting a field value calling condition. For example, FIGS. 5A to 5C show examples of matters related to a natural language generation field among natural language generation criteria, and FIGS. 5D to 5E are natural language generation criteria including importing field values among matters related to all linked words and natural language generation fields. 5G shows a natural language generation criterion including import of a field name among items related to all linked words and natural language generation fields. FIG. 5A shows that a field value (eg, refer to a'imported value' of a generation table) is obtained from a natural language generation field (eg,'total number of generations' corresponding to a field value to be fetched from a generation criterion). 5D shows that the entire connection word is generated according to the natural language generation criterion (e.g., the generation of the entire connection word and the import of the field value) (e.g., the space in the generation table according to the'space' in the generation criterion and the characteristic of the region In the generated table according to the analysis result, refer to'The result of regional characteristic analysis.') and get the field value (e.g., in the natural language generation result corresponding to the administrative code 1171063100, which is a kind of identification ID in the generated table). It shows the results of performing'Seoul Metropolitan City','Songpa-gu','Garak 1-dong', etc. (for example, refer to the'Natural Language Generation Result' field in the generation table). Figure 5f is a whole connection word according to the natural language generation criterion (for example, refer to the natural language generation result in the generation table generated according to the'population ratio analysis result by age group' in the generation criterion) and import the field name (e.g., in the generation criterion). The result of natural language generation corresponding to the administrative code 1171063100, which is a kind of identification ID, in the generated table created according to the'high calling condition' of the field name import, is integrated. The results are shown.

For example, in one example, the field value calling condition may be any one of matching with a specific value of a field, selective matching among two or more specific values, and satisfaction of a specific range of a field value. For example, FIG. 5G shows a calling condition for fetching a field name (eg, a “high calling condition” for fetching a field name in the generation criterion).

Natural language generation field existence check step (S200, S200')

Next, steps S200 and S200' confirming the existence of a natural language generation field will be described with reference to FIGS. 1 to 3. In the steps of confirming the existence of the natural language generation field (S200 and S200'), the processor 10 checks whether at least one or more natural language generation fields according to setting items are present in the read data table. At this time, when the virtual field is fed back in the virtual field generation step (S300, S300') to be described later, in the step of checking whether the natural language generation field exists (S200, S200'), the processor 10 includes the virtual field. Check whether the natural language generation field exists in the data table.

For example, referring to FIG. 2, in one example, in the step of checking whether a natural language generation field exists (S200'), the field to be checked is a natural language generation criterion included in the setting items inputted in the setting item input step (S100'). It can be determined by the items contained in For example, it may be determined by matters related to the natural language generation field included in the natural language generation criteria. Specifically, the field to be checked may be one or more of a field to be retrieved from a field value and a field holding a field value included in a field value calling condition.

Virtual field creation step (S300, S300')

Next, a virtual field generation step (S300, S300') will be described with reference to FIGS. 1 to 4G. In the virtual field generation steps S300 and S300', the processor 10 generates and feeds back at least one virtual field that is a natural language generation field that is not included in the data table read for natural language generation when the natural language generation field does not exist. .

For example, referring to FIG. 2, each of the virtual fields generated in the virtual field generation step (S300') is a field that is generated when the field to be checked does not exist, and processes, processes, or evaluates field values of at least one field of the data table. It holds the field value created by doing so.

In this case, the field value processing may be one of arithmetic processing of numerical field values of two or more fields, standardization of numerical field values in one field, and separation or ranking processing. For example, in this case, the field to be processed for field value may be an existing field or a field including a virtual field generated by processing a previous field value. Field value processing is a reclassification of field values using grouping of at least two or more field values in existing fields or fields created by field value processing, or field values of other numeric fields in the same group as the field value of one field. It may be average to standardized processing for. For example, at this time, one field having the same field value group may be an existing field or a virtual field created by reclassification processing. Field value evaluation is an evaluation date based on the range of numeric field values for one or more fields that have at least a numeric field value in one or more of an existing field, a field generated by field value processing, or a field generated by field value processing. I can.

In this case, field value processing or field value evaluation may be performed according to a virtual field generation criterion included in the setting item in the step of receiving the setting item (S100, S100'). For example, the virtual field generation criterion may be included in the natural language generation criterion or may be provided separately from the natural language generation criterion. For example, FIGS. 4D to 4G illustrate generating a virtual field according to field value processing or field value evaluation according to the virtual field generation criterion. 4D is a reclassification process according to grouping of at least two or more field values in one field according to the virtual field creation criterion (e.g., grouping of'high income class' and'middle class' in the'income class' field in the original table) It represents the creation of a virtual field (for example, the'income class virtual group' field in the generation table). 4G shows another numerical field (eg,'monthly income') of the same group as the field value of one field according to the virtual field generation criterion (e.g., the'high income group and the middle class' group in the'income class virtual group' field in the original table). Field) represents a virtual field (eg,'average monthly income' field in the generation table) generated according to the average processing (eg, monthly income average) of the field value. In Fig. 4g, the virtual field is generated using the virtual field generation criterion representing'income class virtual group' as the dimension group field as the dimension group field,'monthly income' as the dimension view field, and'average' as the field summary criterion. Is shown. 4E shows an evaluation based on a range of numerical field values (eg, field values of the'total households percentile' field in the original table) according to the virtual field generation criterion (eg, the total number of households evaluation criterion) (eg,'in the virtual field generation criterion). It represents the generation of a virtual field (for example, a'total number of households evaluation' field in the generation table) from a field that holds a numeric field value according to the'total number of households evaluation criteria'. 4F shows a range of numerical field values in two fields (e.g., field values of'total households' field and'monthly income evaluation' field in the original table) according to the virtual field generation criteria (e.g., income and household size evaluation criteria) ( For example, according to evaluation based on '9,000' or more for'total number of households' and'high-income area' for'monthly income' in the generation criteria (e.g.,'evaluation of income and number of households' in the virtual field generation criterion) It represents the creation of a field (for example, the'opening strategy for each income and household size evaluation result' field in the generation table).

In addition, field value processing may be performed according to a preset definition for a generated virtual field name or according to a user's setting. During field value processing, for example, average, maximum, minimum, median, maximum frequency, minimum frequency, median frequency, field value percentile, Z score, T score, field value ranking, ranking of field values within each field. Percentiles, separation values (average, median, maximum, minimum, etc.), and separation rates (maximum of separation values by average, median, maximum, minimum, etc. or ratio to the average of field values), etc. It may be calculated according to a preset setting, and operations such as sum, difference, and product of field values corresponding to the same identification ID in two or more fields may be processed according to a preset definition. For example, when field value processing is performed according to a preset definition, a virtual field may be generated by performing field value processing according to a preset definition through analysis of setting items related to a natural language generation field among natural language generation criteria. For example, field value processing may be processed according to a user's setting input.

Figure 4a shows that the virtual field for the'total number of domestic and foreign populations' is generated through the addition operation processing (domestic personage + foreigner personage) for the field values of two fields according to the virtual field generation standard according to the user's setting input. have. FIG. 4B shows the generation of a virtual field (the total number of households percentile) according to the percentile of the field value according to a preset setting. Figure 4c shows a virtual field generation criterion according to the user's setting input (total domestic and foreign population oversufficient = domestic and foreign total population-domestic and foreign total population average) through subtraction operation to create a virtual field for'overs or shortages of total number of domestic and foreigners' It shows that.

For example, in one example, the operation of a numeric field value may be an operator operation using an operator. The operator operation may be an operation such as addition or subtraction, but is not limited thereto. For example, the standardization process of a numerical field value may be a standardization process using a percentile score, a Z score, a T score, and a converted score converted from the corresponding score in one field. For example, the distance processing of the numerical field value may be a separation value using any one of an average value, a maximum value, a minimum value, and a median number as a reference value in one field or a separation degree processing using a distance increase/decrease rate, but is not limited thereto. For example, the ranking processing of the numerical field values may be processing according to any one of an ascending order, a descending order, and an order order on the current data in one field, but is not limited thereto. In this case, a field that is an object of standardization, separation, and ranking processing of a numerical field value may be an existing field or a field generated according to an operation process.

For example, the reclassification processing for the field value may be character processing according to the reclassification or numerical processing according to the reclassification. 4D shows an example of character processing according to reclassification. For example, each of the average or normalization processing during field value processing and field value processing, and numerical processing during reclassification processing for field values may generate a numerical value virtual field. FIG. 4A shows a numerical value virtual field according to an operation process of a field value, for example, a'total number of domestic and foreigners' field, and FIG. 4B shows a numerical value virtual field according to the percentile processing of a field value, for example, a'total number of households percentile' field, FIG. 4C shows a numerical value virtual field according to the processing of the separation of the field value, for example, a field of'oversufficient total number of domestic and foreign nationals'. In addition, FIG. 4G shows a virtual field of numerical values according to average processing during field value processing, for example, a'average monthly income' field. The'average monthly income' field, which is a numerical value virtual field in FIG. 4G, is a summary field (e.g., a summary criterion) for a dimension field, for example, a dimension view field (e.g., a'monthly income' field in the generation criterion) of the'income class virtual group' field. As'average'). For example, in this specification, the dimension field means a field in which the field value is composed of only characters or including characters, and the dimension view field includes a numeric field value corresponding to the identification ID corresponding to the field value of the dimension field. As a field, it is a kind of numeric value field. That is, the dimension view field means a numeric value field having a numeric field value corresponding to the field value of the dimension field.

For example, during reclassification processing for field values, each of character processing and field value evaluation may create a dimension virtual field. 4D shows a dimension virtual field generated according to text processing during reclassification processing, for example, a'income class virtual group' field, and FIGS. 4E to 4F are a dimension virtual field generated according to the field value square, for example,'Evaluation of the total number of households' Field, and'Shopping strategy by income and household size evaluation results' field are shown.

Natural language generation step (S400, S400', S400")

Next, natural language generation steps S400 and S400' will be described with reference to FIGS. 1 to 3 and 5A to 5F. In the natural language generation step (S400, S400', S400"), the processor 10 generates a natural language that analyzes the data table according to the settings in the processor 10 when the natural language generation field exists. In this case, the natural language generation step (S400, S400', S400") includes a whole connection word generation step (S410) and a field value/field name import step (S430, S430'). In the whole connection word generation step (S410), the entire connection word applied to each record of the natural language generation field is generated according to the setting, and in the field value/field name import step (S430), the field of the natural language generation field according to the setting Gets at least one of a value and a field name.

For example, referring to FIG. 2, in one example, in the natural language generation step (S400'), a full connection word applied to the entire record of the natural language generation field is generated according to the natural language generation criterion (S410), and the natural language generation criterion is applied. At least one or more of a field value and a field name of the corresponding natural language generation field may be obtained (S430). For example, in the field value/field name fetching step (S430), at least one of fetching a field value from a natural language generation field and fetching a field name of a natural language generation field meeting a field value calling condition may be performed. For example, FIG. 5A shows a field value (eg, a field value corresponding to selection of a field to be imported in the generation criterion of FIG. 5A) according to a natural language generation criterion (eg, field value import). Refers to'Imported Value' of the generated table). 5B shows two natural language generation fields (e.g.,'income measurement virtual) set in dimension field selection in the generation criterion according to the natural language generation criteria (e.g., bringing together the field value contents of the dimension field and the numerical field value of the dimension view field) This indicates that two field values (e.g., field values such as'high-income and middle-class, 6164157' of the natural language generation result in the generation table) are obtained from the'Group' field and the'average monthly income' field set in the dimension view field selection. have. 5C is a field name matching the field value calling condition according to the field value calling condition (e.g., high calling condition, low calling condition in the generation criteria), which is a natural language generation criterion (for example, administrative code 1171063100, which is a kind of identification ID in the generation table) The field name of the high calling condition imported is '10s, 20s', and the field name of the low calling condition imported is'under 10 years old', etc.). 5D shows that the entire connection word is generated according to the natural language generation criterion (e.g., the generation of the entire connection word and the import of the field value) (e.g., the space in the generation table according to the'space' in the generation criterion and the characteristic of the region In the generated table according to the analysis result, refer to'The result of regional characteristic analysis.') and get the field value (e.g., in the natural language generation result corresponding to the administrative code 1171063100, which is a kind of identification ID in the generated table). It shows the results of performing'Seoul Metropolitan City','Songpa-gu','Garak 1-dong', etc. (for example, refer to the'Natural Language Generation Result' field in the generation table). Figure 5e shows an entire connection word according to the natural language generation criterion (e.g., refer to the natural language generation result in the generation table generated according to the'regional overall average' in the generation criterion,'by name') and the field value import (e.g., generation The results of the integration are shown in the results of natural language generation in the generation table generated according to the'total number of domestic and foreigners oversufficient' and'overs or shortages of the total number of domestic and foreigners' in the standard. Figure 5f is a whole connection word according to the natural language generation criterion (for example, refer to the natural language generation result in the generation table generated according to the'population ratio analysis result by age group' in the generation criterion) and import the field name (e.g., in the generation criterion). The result of natural language generation corresponding to the administrative code 1171063100, which is a kind of identification ID, in the generated table created according to the'high calling condition' of the field name import, is integrated. The results are shown.

For example, referring to FIGS. 5D to 5F, in the steps of generating natural language (S400, S400', S400"), natural language is generated according to a processing priority included in the natural language generation criteria.

For example, referring to FIGS. 5D to 5F, the steps of generating natural language (S400, S400', S400") are a process of changing lines for the next sentence when the generated natural language sentence is terminated (e.g., ' (Refer to the natural language generation result in the generation table according to the'line break period') can be included.

For example, referring to FIG. 3, in an example, when there are a plurality of natural language generation fields and a plurality of virtual fields are generated, a process of generating a natural language for a virtual field generated after one virtual field may be repeated. .

In addition, referring to FIG. 3, in an example, when there are a plurality of natural language generation fields and a plurality of virtual fields are generated, a plurality of natural language generation fields including a plurality of virtual fields generated after the generation of a plurality of virtual fields are Can generate natural language.

Next, an example of an electronic device that generates natural language from a data table according to another embodiment of the present invention will be described. In this case, it can be understood with reference to embodiments of the natural language generation method from the above-described data table and FIGS. 1 to 5F.

FIG. 6 is a schematic diagram illustrating an electronic device that generates natural language from a data table according to an embodiment of the present invention, and FIG. 7 is a schematic diagram illustrating a natural language generation from a data table according to an embodiment of the present invention. It is a schematic configuration diagram showing a processor of an electronic device.

Referring to FIG. 6, an electronic device 1 for generating natural language from a data table according to an example is an electronic device 1 including at least one processor 10. At this time, the at least one processor 10 receives setting items to generate a natural language for analyzing a data table including a plurality of fields, and determines whether at least one or more natural language generation fields according to the settings are present in the read data table. Confirm. The natural language generation field is a field that generates natural language. In this case, the field of the data table includes a field name and at least one field value. Referring to FIG. 6, setting items are input from a user through the input device 20, and the processor 10 transmits or receives the setting items from the input device 20. For example, the processor 10 of FIG. 6 reads a data table to generate a natural language. For example, it is checked whether at least one or more natural language generation fields according to settings transmitted or received from the processor 10 of FIG. 6 to the natural language generation field verification module 11 of FIG. 7 exist in the read data table. When the natural language generation field does not exist, the at least one processor 10 generates and feeds back at least one virtual field that is a natural language generation field that is not included in a data table read for natural language generation. For example, the processor 10 of FIG. 6 to the virtual field generation module 13 of FIG. 7 generate and feed back a virtual field. In the case of feedback of a virtual field, the at least one processor 10 checks whether a natural language generation field exists in a data table including the virtual field. For example, in the case of the feedback of the virtual field in the processor 10 of FIG. 6 to the natural language generation field verification module 11 of FIG. 7, the existence of the natural language generation field is checked. At least one processor 10 generates a natural language that analyzes the data table according to the settings when the natural language generation field exists, but generates and sets the entire connection word applied to the entire record of the natural language generation field according to the settings. A natural language is generated by including a process of obtaining at least one of a field value and a field name of a natural language generation field according to the item. For example, the processor 10 of FIG. 6 to the natural language generation module 15 of FIG. 7 generate natural language.

For example, referring to FIG. 6, the electronic device 1 may include a memory 30. The memory 30 may temporarily store the data table read by the processor 10 or when the processor 10 reads the data table from the external database 3. In addition, the memory 30 may store setting items input from the input device 20 by the user.

In one example, the setting item input from the processor 10 may include a natural language generation criterion for generating a natural language. In this case, the natural language generation criterion may include matters about the entire linked language and matters about the natural language generation field. The matters regarding the natural language generation field may include at least one of fetching a field value from the natural language generation field and fetching a field name of a natural language generation field meeting a field value calling condition. For example, in the processor 10 or the natural language generation field verification module 11, the verification target field, which is a natural language generation field that is to be checked for existence, is a field to which the field value is fetched, and the field value included in the field value calling condition. It may be any one or more of the owned fields.

For example, in one example, the processor 10 of FIG. 6 to the virtual field generation module 13 of FIG. 7 process, process, or evaluate field values of at least one field of the data table when the field to be checked does not exist. You can create a virtual field that holds the generated field value. In this case, the field value processing may be one of arithmetic processing of numerical field values of two or more fields, standardization of numerical field values in one field, and separation or ranking processing. Field value processing is a reclassification of field values using grouping of at least two or more field values in existing fields or fields created by field value processing, or field values of other numeric fields in the same group as the field value of one field. It may be average to standardized processing for. Field value evaluation is an evaluation date based on the range of numeric field values for one or more fields that have at least a numeric field value in one or more of an existing field, a field generated by field value processing, or a field generated by field value processing. I can. Field value processing or field value evaluation may be performed according to the virtual field generation criterion included in the input settings. Field value processing may be performed according to a predefined definition of a virtual field name to be created or according to a user's setting.

For example, in one example, the operation of a numeric field value may be an operator operation using an operator. The standardization process of the numerical field value may be a standardization process using a percentile score, a Z score, a T score, and a converted score converted from the corresponding score in one field. The separation degree processing of the numerical field value may be a separation degree processing using one of an average value, a maximum value, a minimum value, and a median number as a reference value or a separation degree increase/decrease rate in one field. The ranking processing of the numerical field values may be processing according to any one of an ascending order, a descending order, and an order order on the current data in one field. Fields subject to standardization, separation, and ranking processing of numerical field values may be existing fields or fields generated according to calculation processing. Reclassification processing for field values may be character processing according to reclassification or numerical processing according to reclassification. The processor 10 of FIG. 6 to the virtual field generation module 13 of FIG. 7 generate a numerical value virtual field according to the average or standardization processing during field value processing and field value processing, and numerical processing during reclassification processing for field values. Can be generated. The processor 10 of FIG. 6 to the virtual field generation module 13 of FIG. 7 may generate a dimension virtual field according to character processing and field value evaluation during reclassification processing for field values.

In addition, in an example, the field value calling condition may be any one of matching with a specific value of a field, selective matching among two or more specific values, and satisfaction of a specific range of a field value. The processor 10 of FIG. 6 to the natural language generation module 15 of FIG. 7 may generate natural language according to a processing order included in the natural language generation criterion. The processor 10 of FIG. 6 to the natural language generation module 15 of FIG. 7 may perform a process of changing lines for the next sentence when the generated natural language sentence is terminated.

In another example, the processor 10 of FIGS. 6 to 7 performs a process of generating a natural language for a virtual field generated after generating one virtual field when there are a plurality of natural language generation fields and a plurality of generated virtual fields. Can repeat. Or, in another example, the processor 10 of FIGS. 6 to 7 may include a plurality of natural language generation fields and a plurality of virtual fields generated after generation of a plurality of virtual fields, when there are a plurality of natural language generation fields and a plurality of generated virtual fields. Natural language can be generated for the generated field.

[Drawing Example]

Next, a specific embodiment will be described with reference to FIGS. 3 to 5F.

Referring to FIG. 3, when a field for generating natural language exists after obtaining data of an original field to be analyzed, a natural language is generated to derive a natural language generation result. When performing natural language generation, functions such as full connection word, field value import, field name import, and line break cycle can be performed. In addition, when there is no field to generate natural language in the original data table, virtual field is created by creating a virtual field for natural language generation. Virtual field creation can be performed by creating a virtual numeric field, creating a virtual separation field, creating a virtual dimension field, and creating a virtual dimension summary field. In this case, the virtual numeric field, the virtual separation field, and the virtual dimension summary field may be a numeric value virtual field, and the virtual dimension field may be a virtual field when the field value is a character field value or a mixed field value mixed with numbers and characters. For example, the virtual field generation may be performed according to the virtual field generation condition included in the natural language generation condition or provided separately. Alternatively, a virtual field may be generated according to a predefined definition according to a field name generated according to a natural language generation condition. When a virtual field required by the user is generated, it is determined that a field for natural language generation exists, and the natural language is performed in the same manner as the natural language generation process when the field exists. A field includes a field name and at least one field value in the data, and the at least one field value corresponds to an identification ID, respectively. A field does not mean only one in the data table, but includes multiple fields. If there is an additional natural language to be generated after the natural language is generated, it is repeatedly performed, and finally, when there is no additional natural language to be generated, it is terminated. Natural language generation is not only sequentially performing virtual field creation and natural language generation, but also performing all the virtual field creation operations for natural language generation, and then using them to perform both natural language generation, that is, in two steps. Includes performing.

Referring to FIG. 3, in performing virtual field creation, creating a virtual numeric field using an operator for a field, creating a virtual separation field regarding how far apart is in the field distribution, creating a virtual dimension field such as grouping dimension fields, and creating a dimension field. The creation of a virtual dimension summary field such as a summary can be optionally included. For example, creating a virtual numeric field can optionally include creating a virtual numeric field using a speaker, creating a virtual numeric field using a standard index, and creating a virtual numeric field using alignment, and creating a virtual separation field is a virtual separation value using an operator. Creating a field, creating a virtual separation rate field using a speaker, etc. can be optionally included, and creating a virtual dimension field includes creating a virtual dimension field of a dimension field, creating a virtual dimension field of a single numeric field, creating a virtual dimension field of multiple numeric fields, The creation of a virtual dimension field of a plurality of characters and a numeric hybrid field may be optionally included.

Also, referring to FIG. 3, an example of performing natural language generation is shown. To perform natural language generation, a linkage word for each record is not given, but a global linkage applied to all records, a field value that gets a field value corresponding to each record, and a field when a field value corresponding to each record meets the conditions. It may include fetching the field name for fetching the name, and the line break cycle for changing the line for the next sentence when the natural language sentence is terminated. For example, natural word generation can be performed based on the import of the entire connection word and the field value or the import of the field name. Global concatenation consists of all concatenation words, and field value import can include numeric field value import, character field value import, dimension view field value, etc., field name import consists of field name import. And, the line break cycle can be composed of a line break cycle.

4A shows an example of creating a virtual numeric field using an operator while creating a virtual numeric field in performing virtual field creation in the present invention. A virtual numeric field can be created using four arithmetic operations on one or more fields by creating a separate virtual numeric field by using an operator on the original field value. This figure is an example of the process of creating a virtual field for the total number of domestic and foreign populations by adding endurance and foreigners. In addition to the four arithmetic symbols, such as +-× ÷, the operator includes general symbols for operations such as = () root square. The virtual field created through this process, that is, the generation table, may be composed of the identification ID of the original field, the virtual field name defined by the user, and the value of the virtual field.

4B is a diagram showing an example of creating a virtual numeric field using a standard index during virtual numeric field creation in performing virtual field creation according to the present invention. Creating a virtual numeric field using a standard index is to determine the location of a corresponding record among the distribution of field values across the record, and a percentile index can be used. This figure is an example of the process of calculating the percentile index for the total number of households. When calculating the standard index, it is used as a percentile, but includes statistical indexes for standardization such as standardized Z score, which is a standardized index. The virtual field created through this process, that is, the generation table, may be composed of the identification ID of the original field, the virtual field name defined by the user, and the value of the virtual field.

4C is a diagram showing an example of creating a virtual separation value field using an operator during virtual separation field creation in performing virtual field creation according to the present invention. Creating a virtual separation value field using an operator may be to determine how many values are separated by a specific record compared to the average field value of the entire record. The average is used as the reference value of the separation, but includes the maximum value, minimum value, and median, and the reference value of the separation can be calculated by an internal algorithm without a separate operator. This drawing is an example of a process of calculating an excess or shortage of a field value of the total number of foreigners and foreigners of a specific record compared to the average of the field value of the total number of foreigners and foreigners of all records. The virtual field created through this process, that is, the generation table, may be composed of the ID of the original field, the name of the virtual field defined by the user, and the value of the virtual field.

4D is a diagram illustrating an example of creating a virtual dimension field of a dimension field during virtual dimension field creation in performing virtual field creation according to the present invention. Creating a virtual dimension field of a dimension field is used when it is used to analyze a dimension from a different perspective from the current dimension point of view, and creating a virtual dimension field of a dimension field creates a virtual dimension field by grouping dimensions within one field. As a result, the dimension can include all of the numeric type, character type, and mixed type of numeric characters. In this drawing, in the original income class field, what consists of high-income, middle-class, general and low-income class is changed to "high-income class" and "middle class" value by integrating "high-income class" and "middle class", and "general class", "low-income class" Is an example of the process of virtually configuring the same. The virtual field created through this process, that is, the generation table, may be composed of the identification ID of the original field, the virtual field name defined by the user, and the value of the virtual field. Creating a virtual dimension field of a dimension field involves creating a virtual dimension field for not only one dimension field of the original, but also two or more dimension fields. The virtual field created through this process, that is, the generation table, may be composed of the ID of the original field, the name of the virtual field defined by the user, and the value of the virtual field.

4E is a diagram illustrating an example of creating a virtual dimension field of a single numeric field during virtual dimension field creation in performing virtual field creation according to the present invention. Creating a virtual dimension field of a single numeric field is used to classify a numeric value of one field into a dimension according to a range of numeric conditions. To determine whether the value of a single numeric field for each record is within the numerical condition range, the start value and the end value of a single numeric field can be set, and whether the start value and the end value are included can be defined with equal signs and inequality signs. Equal sign and inequality sign can be set with symbols such as = <> ≠ ≤ ≥. In this drawing, if the value of the total number of households percentile field of each record is less than 10%, it is "very small." , If it is more than 10% and less than 35%, it is “little”, if it is more than 35% and less than 65%, it is “normal”, if it is more than 65% and less than 90%, it is “a lot”, and if it is more than 95% and “very much” is." This is an example of the process of calculating the total number of households evaluation result for each record after it is defined as. The virtual field created through this process, that is, the generation table, may be composed of the identification ID of the original field, the virtual field name defined by the user, and the value of the virtual field.

4F is a diagram showing an example of creating a virtual dimension field of a plurality of character-numeric hybrid fields while creating a virtual dimension field in performing virtual field creation according to the present invention. Creating a virtual dimension field of multiple character value hybrid fields is used to create a virtual dimension for a field in which two or more characters and numbers are mixed. To create a virtual dimension field of multiple character-value hybrid fields, you can select two or more fields to use, and the inclusion relationship is defined as an inequality sign with an inequality sign for the start value and the end value of each selected field, and the field value of each record is Classified when included in the range of the start value and the end value. The inequality sign can be set with a symbol such as = <> ≠ ≤ ≥.

Figure 4f is an example written based on the total number of households and monthly income evaluation field value of each record. To explain this, if the total number of households is 9,000 or more and the monthly income evaluation field value is a high-income area, "The area with a large number of high-income households is given priority to opening stores of middle- and high-end brands If the total number of households is less than 9,000 and the monthly income evaluation field value is high-income area, it is considered that there is no need to open a store in a high-income area or because the number of households is small.”, When the total number of households is 9,000 or more and the monthly income evaluation field value is low-income If it is a region, "It is necessary to first consider opening a low-cost brand store because of the large number of low-income regions or households.", If the total number of households is less than 9,000 and the monthly income evaluation field value is low-income region, then, "It is a low-income region and has a small number of households. This is an example of the process of calculating the evaluation result for each record after definition as "I don't think there is a need to do it." In this drawing, although the description is made based on two fields, it includes creating a virtual dimension field for three or more multi-character numeric hybrid fields. The virtual field created through this process, that is, the generation table, may be composed of the identification ID of the original field, the virtual field name defined by the user, and the value of the virtual field.

4G is a diagram showing an example of creating a virtual summary field in performing virtual field creation according to the present invention. Creating a virtual dimension summary field is used when a field to be analyzed based on the dimension field, that is, a view field value, is to be summarized and analyzed. Creating a virtual dimension summary field consists of selecting a field to be analyzed, that is, a field of view, and setting a summary criterion for the value of the selected field of view after selecting the dimension field. Average, minimum, maximum, median, etc. can be used as the summary criterion for the field of view. In this drawing, "income class virtual group" was selected as the dimension group field, and "monthly income" as the dimension view field, and the summary standard of the dimension view field was defined as "average". This is an example. The virtual field created through this process, that is, the generation table, may be composed of the identification ID of the original field, the virtual field name defined by the user, and the value of the virtual field.

Although the creation of the virtual field in FIGS. 4A to 4G has been presented to explain the case in detail, it is obvious that each section can be integrated. For example, creating a virtual numeric field may be performed by integrating at least two or more of creating a virtual numeric field using an operator, creating a virtual numeric field using a standard index, and creating a virtual numeric field using alignment. Creating a virtual field in FIGS. 4A to 4G is an example of creating a field for each section in order to describe the case in detail, and it is an example that several virtual field tables are configured, but an identification ID and a user are in one virtual table. It may include creating in a form that includes all the virtual fields defined by.

5A is a diagram showing an example of obtaining a value of a numeric field in performing natural language generation according to the present invention. Get numeric field value can be used to generate natural language by getting the field value of a specific record. In this drawing, it is an exemplary diagram of a process of generating natural language by registering to get the value of the total number of households field and getting the value of the total number of households field for each record. The natural language generated through this process, that is, the generation table, may be composed of an identification ID of an original field and a natural language generation result according to a user-defined criterion.

5B is a diagram showing an example of obtaining a dimension view field value in performing natural language generation according to the present invention. Get the dimension view field value can be used when you want to get the dimension view field value that matches the dimension content of the dimension field. Dimension view field value import consists of dimension field selection, dimension field content, dimension view field selection, value to be imported, and insertion characters when importing more than two, and the insertion character is a connection defined by the user with special symbols such as commas (, ). You can use sentences, etc. In this drawing, for the income class virtual group dimension and the average monthly income view field in the original data, the dimension field selection is “income class virtual group”, the dimension field content is “high income and middle class”, and the dimension view field selection is in the income class virtual group. In the case of high-income and middle-class, the high-income, middle-class, and average monthly income are imported, and when two or more are imported, a comma (,) is displayed to create natural language. The natural language generated through this process, that is, the generation table, may be composed of an identification ID of an original field and a natural language generation result according to a user-defined criterion.

5C is a diagram showing an example of getting a field name in performing natural language generation according to the present invention. Field name import is to import field names for each record when the field value of a specific record matches user conditions. Not only one field name is imported, but when there are two or more fields that meet user conditions for each field. Also includes bringing. When there are several fields that satisfy the user condition, characters such as commas (,) can be inserted between the field name and the field name. In this case, when importing two or more field names, it is executed by the insert character command. In this drawing, in the case of “High” for fields under 10, 10, 20, and 30 as a high calling condition, “High” for fields under 10, 10, 20, and 30 as a low calling condition. In the case of low”, a natural language is generated for each record, and when two or more field names are generated, a comma (,) is displayed in the inserted character command. The natural language generated through this process, that is, the generation table, may be composed of an identification ID of an original field and a natural language generation result according to a user-defined criterion.

5A to 5C are presented to specifically explain the case of natural language generation, but it is obvious that each section can be integrated and created. For example, the field value can be imported by integrating at least two of the field value import, the numeric field value import, the character field value import, and the dimension view field value import.

5D is a diagram showing an example of generating a natural language using a field value import, an entire connection word, and a line break period during execution using natural language generation according to the present invention. This figure shows the field value for the primary field for each record, the entire space linkage, the field value for the secondary field, the entire space linkage, the field value for the administrative name field, the space linkage, This is the result of characteristic analysis.” This is an example of the process of generating natural language by sequentially applying the entire connection word and line break cycle. The natural language generated through this process, that is, the generation table, may be composed of an identification ID of an original field and a natural language generation result according to a user-defined criterion.

5E is a diagram showing an example of natural language generation using an entire connection word, a field value import, and a line feed period during natural language generation according to the present invention. In this drawing, the field for evaluating the total number of domestic/foreigners oversufficiency may be a field generated by using the total number of domestic/foreigners field to generate a field value of the total number of domestic/foreigners oversufficiency fields created in the process of creating a virtual numeric field using a single numeric virtual dimension condition. This drawing is for each record, “total concatenation word, space total concatenation word, “total population and shortage” field values for each record, “total concatenation word, space total concatenation word, “total domestic and foreign population surplus and shortage evaluation” for each record. This is an example of the process of generating natural language by sequentially applying field value and line break cycles. The natural language generated through this process, that is, the generation table, may be composed of an identification ID of an original field and a natural language generation result according to a user-defined criterion.

FIG. 5F is a diagram showing an example of natural language generation using an entire connection word, a field name import, and a line break period during natural language generation according to the present invention. In this drawing, the evaluation result for each age group may be a field generated by creating a single numerical virtual field for the percentile field for each age group. This drawing relates to an example of generating a natural language when an evaluation field value for each age group of each record corresponds to a field name call. This drawing shows the results of the evaluation by age group for each record, “Results of Analysis of Population Ratio by Age Group” All concatenation words, “Space” total concatenation words, “High call condition” field names import, “Space” all concatenation words, “Low call condition” Get field name, “Population ratio is low.” This is an example of the process of generating natural language by sequentially applying the whole connection word and the line break period. The natural language generated through this process, that is, the generation table, may be composed of an identification ID of an original field and a natural language generation result according to a user-defined criterion.

The natural language generation results of FIGS. 5D to 5F may be combined to generate a natural language by combining at least two.

In the above, the above-described embodiments and the accompanying drawings are not intended to limit the scope of the present invention, but are illustratively described to help those of ordinary skill in the art understand the present invention. That is, various embodiments of the present invention may be implemented in various modified forms according to various combinations of the above-described components without departing from the essential characteristics of the present invention. Accordingly, the scope of the present invention should be interpreted according to the invention described in the claims, and includes not only the above-described embodiments, but also various modifications, alternatives, and equivalent embodiments by those of ordinary skill in the art.

1: electronics 3: database
10: processor 11: natural language generation field verification module
13: Virtual field generation module 15: Natural language generation module
20: input device 30: memory

Claims (14)

Receiving, by a processor, setting items to generate a natural language for analyzing a data table including a plurality of fields;
Checking whether at least one or more natural language generation fields according to the settings are present in the read data table, the natural language generation field being a field to generate the natural language;
Generating and feeding back at least one virtual field that is the natural language generation field that is not included in the data table read for natural language generation when the natural language generation field does not exist; And
In the presence of the natural language generation field, the processor generating a natural language for analyzing the data table according to the setting item,
In the step of checking whether the natural language generation field exists in the case of the feedback of the virtual field, it is checked whether the natural language generation field exists in the data table including the virtual field,
The step of generating the natural language is:
Generating an entire connection word applied to the entire record of the natural language generation field according to the setting item; And
Including the step of obtaining at least one or more of a field value and a field name of the natural language generation field according to the setting item,
The field includes the field name and at least one field value,
In the step of receiving the setting item, the setting item includes a natural language generation criterion for generating the natural language, and the natural language generation criterion includes a matter on the entire linked language and a matter on the natural language generation field,
The matters regarding the natural language generation field include at least one of fetching a field value from the natural language generation field and fetching a field name of the natural language generation field meeting a field value calling condition,
In the step of checking whether the natural language generation field exists, the verification target field is at least one of a field to be retrieved from the field value and a field holding a field value included in the field value calling condition,
Each of the virtual fields is a field generated when the field to be checked does not exist, and holds a field value generated by processing, processing, or evaluating field values of at least one field of the data table,
The field value processing is one of arithmetic processing of numerical field values of two or more fields, standardization of the numerical field values in one field, and separation or ranking processing,
The field value processing is performed by reclassifying the field value using grouping of at least two or more field values in the existing field or the field generated by the field value processing, or another numerical value in the same group as the field value of one field. It is the average or standardized processing of the field value of the field,
The field value evaluation is the numerical field value for one or more fields holding at least the numerical field value in one or more of the existing field, a field generated by processing the field value, and a field generated by processing the field value. Is an evaluation based on the scope of
The field value processing to the field value evaluation are performed according to a virtual field generation criterion included in the setting item in the step of receiving the setting item,
The field value processing is performed according to a preset definition for a generated virtual field name or a user's setting.
delete delete In claim 1,
The operation processing of the numerical field value is an operator operation using an operator,
The standardization process of the numerical field value is a standardization process using a percentile score, a Z score, a T score, and a converted score converted from the corresponding score in one field,
The separation degree processing of the numerical field value is a separation degree processing using a separation degree value or a separation degree increase/decrease rate using any one of an average value, a maximum value, a minimum value, and a median number as a reference value in one field,
The ranking processing of the numerical field values is processing according to any one of an ascending order, a descending order, and an order order on the current data in one field,
A field that is the target of the standardization, separation, and ranking processing of the numerical field value is an existing field or a field generated according to the calculation process,
Reclassification processing for the field value is character processing according to reclassification or numerical processing according to reclassification,
Each of the numerical processing during the field value processing and the average or normalization processing during the field value processing and the reclassification processing for the field value creates a numerical value virtual field,
Each of the character processing and the field value evaluation during the reclassification processing of the field value generates a dimension virtual field.
In any one of claims 1 or 4,
The field value calling condition relates to any one of matching with a specific value of a field, selective matching of two or more of the specific values, and satisfaction of a specific range of a field value,
In the step of generating the natural language, the natural language is generated according to the processing priority included in the natural language generation criterion,
The generating of the natural language includes a process of changing a line for a next sentence when the generated natural language sentence is terminated.
In any one of claims 1 or 4,
When the number of natural language generation fields is plural and the virtual fields to be generated are plural, natural language generation from a data table, characterized in that the process of generating the natural language for the virtual field generated after generating one virtual field is repeated. Way.
In any one of claims 1 or 4,
When there are a plurality of natural language generation fields and a plurality of virtual fields are generated, the natural language is generated for a plurality of the natural language generation fields including the plurality of virtual fields generated after the generation of the plurality of virtual fields. How to generate natural language from a data table.
In the electronic device including at least one processor,
The processor,
In order to generate a natural language for analyzing a data table including a plurality of fields, setting items are input, and it is checked whether at least one or more natural language generation fields according to the setting items exist in the read data table, and the natural language generation field is It is a field that is a target for generating the natural language,
In the absence of the natural language generation field, at least one virtual field that is the natural language generation field not included in the data table read for natural language generation is generated and fed back, and in the case of the feedback of the virtual field, the virtual field is included. Check whether the natural language generation field exists in the data table,
In the presence of the natural language generation field, a natural language for analyzing the data table is generated according to the setting details, and the process of generating a global connection word applied to the entire record of the natural language generation field according to the setting details and the setting details Generating the natural language, including the process of obtaining at least one or more of a field value and a field name of the natural language generation field according to,
The field includes the field name and at least one field value,
The setting information received from the processor includes a natural language generation criterion for generating the natural language, and the natural language generation criterion includes a matter on the entire linked language and a matter on the natural language generation field, and the natural language generation field Regarding matters includes at least one or more of fetching a field value from the natural language generation field and fetching a field name of the natural language generation field meeting a field value calling condition,
The verification target field, which is the natural language generation field to be checked for existence in the processor, is at least one of a field to be retrieved from the field value and a field to hold a field value included in the field value calling condition,
The processor generates the virtual field that holds field values generated by processing, processing, or evaluating field values of at least one field of the data table when the field to be checked does not exist,
The field value processing is one of arithmetic processing of numerical field values of two or more fields, standardization of the numerical field values in one field, and separation or ranking processing,
The field value processing is performed by reclassifying the field value using grouping of at least two or more field values in the existing field or the field generated by the field value processing, or another numerical value in the same group as the field value of one field. It is the average or standardized processing of the field value of the field,
The field value evaluation is the numerical field value for one or more fields holding at least the numerical field value in one or more of the existing field, a field generated by processing the field value, and a field generated by processing the field value. Is an evaluation based on the scope of
The field value processing to the field value evaluation are performed according to the virtual field creation criterion included in the input settings,
The field value processing is performed according to a predefined definition of a virtual field name to be generated or according to the user's setting.
delete delete In claim 8,
The operation processing of the numerical field value is an operator operation using an operator,
The standardization process of the numerical field value is a standardization process using a percentile score, a Z score, a T score, and a converted score converted from the corresponding score in one field,
The separation degree processing of the numerical field value is a separation degree processing using a separation degree value or a separation degree increase/decrease rate using any one of an average value, a maximum value, a minimum value, and a median number as a reference value in one field,
The ranking processing of the numerical field values is processing according to any one of an ascending order, a descending order, and an order order on the current data in one field,
A field that is the target of the standardization, separation, and ranking processing of the numerical field value is an existing field or a field generated according to the calculation process,
Reclassification processing for the field value is character processing according to reclassification or numerical processing according to reclassification,
The processor generates a numerical value virtual field according to each of the average or standardized processing during the field value processing and the field value processing, and the numerical processing during the reclassification processing for the field value, and the reclassification processing for the field value And generating a dimension virtual field according to each of the character processing and evaluation of the field value.
In any one of claims 8 or 11,
The field value calling condition relates to any one of matching with a specific value of a field, selective matching of two or more of the specific values, and satisfaction of a specific range of a field value,
The processor generates the natural language according to the processing priority included in the natural language generation criterion,
The processor is an electronic device for generating natural language from a data table, characterized in that when the generated natural language sentence is terminated, the process of changing a line for a next sentence is performed.
In any one of claims 8 or 11,
The processor, when the number of natural language generation fields is plural and the virtual fields to be generated are plural, repeats the process of generating the natural language for the virtual field generated after one virtual field is generated. An electronic device that performs natural language generation from
In any one of claims 8 or 11,
The processor generates the natural language for a plurality of the natural language generation fields including the plurality of virtual fields generated after the generation of the plurality of virtual fields when the number of natural language generation fields is plural and the generated virtual fields are plural. An electronic device for generating natural language from a data table.

Images