JP2010102643A - Apparatus and method for interactive guidance support - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve efficiency of interactive guidance and give guidance with uniform content by automatically determining a behavior modification stage without imposing a burden on a person concerned. <P>SOLUTION: An interactive guidance support apparatus includes: a speech recording acquisition unit for acquiring speech record data in which a dialogue between a guide and the person concerned in the interactive guidance is recorded; a decomposition unit for decomposing the dialogue contained in the speech record data in units of sentences; a continuity evaluation unit for evaluating the continuity between two sentences for all combinations of sentences obtained by the decomposition unit; a feature value calculation unit for calculating a degree of effect which is a feature value representing the intensity of effect given to other sentences and a degree of association which is a feature value representing the intensity of association with other sentences, based on the continuity obtained by the continuity evaluation unit, for each sentence, respectively; a distribution map creation unit for mapping the feature values of the sentences spoken by the person concerned in the space of the feature values consisting of the degrees of effect and association; and a decision part for determining the behavior modification stage of the person, based on a distribution pattern of the feature values in the space of the feature values. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a technique for supporting interactive guidance such as health guidance.

  From April 2008, a specific screening / specific health guidance system was started as a measure against metabolic syndrome (visceral fat syndrome). It was obliged to take the period. However, while the number of people who are subject to specific health guidance is expected to reach about 20 million, there are only about 40,000 public health nurses who provide health guidance. Moreover, it has been pointed out that the number of health nurses with behavioral science skills is very limited.

  In order to obtain high effectiveness through interactive instruction, the instructor correctly understands the current behavior change stage (consciousness level) of the instructor and gives the appropriate instruction to the target according to the stage. It is necessary to promote behavioral change for improvement. The following is an example of a behavioral transformation stage model based on behavioral psychology.

(1) Indifference period: A state where there is no intention to change behavior within 6 months.
Guidance is needed to raise awareness of the need.
(2) Period of interest: A state where there is a willingness to change behavior within six months.
Guidance for motivation and confidence is necessary.
(3) Preparatory period: A state in which there is a willingness to change behavior within one month.
Need to commit action plan.
(4) Action period: State within 6 months after changing action.
It is necessary to give reward and follow-up.
(5) Maintenance period: A state of 6 months or more after changing behavior.

  However, it is extremely difficult to objectively specify the behavior change stage from the conversation with the target person, and there is a problem that the content of the instruction varies greatly depending on the experience and skills of the individual leader.

Therefore, conventionally, it has been generally performed to specify the level of consciousness of the target person through a questionnaire. Patent Document 1 proposes a device that displays a question for investigating / confirming the consciousness level on the screen, causes the subject to input an answer, determines the consciousness level based on the content of the answer, and outputs advice. ing. This device corresponds to an electronic version of a handwritten questionnaire. However, such a method using a questionnaire is inefficient because it requires many questions only to determine the level of consciousness. In addition, it is not preferable to conduct a questionnaire because a burden is generated on both the leader and the target person.
JP 2002-358372 A

  The present invention has been made in view of the above circumstances, and the purpose of the present invention is to improve the efficiency of interactive guidance by enabling automatic determination of the behavioral transformation stage without imposing a burden on the subject. The goal is to homogenize the content of the instruction.

  In order to achieve the above object, the present invention adopts the following configuration.

  An interactive instruction support device according to the present invention includes an utterance record acquisition unit that acquires utterance record data in which conversation between a leader and a subject in interactive instruction is recorded, and conversations included in the utterance record data in sentence units. The continuity between the decomposition part to be decomposed and the two sentences is evaluated for all combinations of sentences obtained by the decomposition part, and the continuity obtained by the continuity evaluation part. Based on this, a feature amount calculation unit that calculates an influence degree that is a feature amount that represents the strength of influence on other sentences and a relevance degree that is a feature amount that represents the strength of association with other sentences for each sentence. And a distribution map creation unit that maps the feature amount of the sentence spoken by the subject to the feature amount space composed of the influence degree and the relevance degree, and based on the distribution pattern of the feature amount in the feature amount space, Judgment to determine the behavior change stage And, equipped with a.

  According to this configuration, the behavior change stage (consciousness level) of the subject can be automatically determined from the utterance record data in which the conversation between the leader (an expert such as a public health nurse) and the subject is recorded. Therefore, the instructor can efficiently concentrate on the conversation with the target person during the face-to-face instruction. In addition, since the behavior change stage of the target person is automatically and objectively determined, it is possible to make the content of the guidance uniform regardless of the experience and skills of the leader.

  Here, the continuity evaluation unit has in advance an evaluation rule including a determination condition that defines a combination of the first character string and the second character string, and the first sentence of the two sentences is In the case where the first character string is included and the second sentence includes the second character string, it is preferable to determine that the two sentences have continuity.

  Thereby, it becomes possible to evaluate the continuity between sentences (it can also be called a causal relationship) by simple processing.

  It further includes a determination condition generation unit that automatically generates a determination condition for the evaluation rule using learning data, and the learning data is composed of a pair of a first sentence and a second sentence having continuity. The determination condition generation unit identifies the type of each sentence included in the learning data based on the description included in the sentence, and determines the first sentence type and the first sentence from the learning data. The combination of the two sentence types is the same, and a pair in which the same or similar word is included in each of the first sentence and the second sentence is extracted, and from each extracted pair, It is preferable to generate a determination condition in which the description or / and the word in one sentence is the first character string, and the description or / and the word in the second sentence is the second character string.

  As a result, it is possible to automatically generate an evaluation rule used for continuity evaluation, and it is possible to accurately determine the continuity of a conversation sentence by using the evaluation rule generated in this way.

  Conversational sentences are abbreviated and anaphoric, and contain interjections and inversions. Therefore, only general morphological analysis (words) and syntactic analysis (such as dependency) require context (ie, sentence continuity). It is difficult to evaluate. On the other hand, in the present invention, (1) the type of sentence can be identified from the end of the conversation sentence, etc. (2) the combination of sentence types that can be continued is determined, (3) the two sentences are If there is a relationship between the combinations of types of sentences that can be continued, and there is a relationship between a word contained in one sentence and a word contained in the other sentence, it can be considered that two sentences are consecutive. Based on the idea, the above-described determination conditions and evaluation methods are adopted.

That is, the process of extracting a pair “the combination of the first sentence type and the second sentence type is the same” is performed by selecting (2) “successive combination of sentence types” from a large number of learning data. Is equivalent to detecting. In addition, the process of “extracting a pair containing the same or similar word in each of the first sentence and the second sentence” is a word that frequently occurs in the first sentence and the second sentence. This is equivalent to detecting a combination (including synonyms). The high frequency of co-occurring in two sentences can be considered to be related to the two words. Therefore, the evaluation is performed using the “determination condition in which the description or / and the word in the first sentence is the first character string, and the description or / and the word in the second sentence is the second character string”. The process is as described in (3) above: “Two sentences are in a relationship of combinations of sentence types that can be continued, and there is a relationship between a word contained in one sentence and a word contained in the other sentence”. It is equivalent to the process of evaluating whether or not.

  The determination unit includes a template that defines a distribution pattern corresponding to each behavior modification stage in advance, and identifies a corresponding behavior modification stage by comparing the distribution pattern of the feature amount in the feature amount space with each template. It is preferable to do.

  Preferably, guidance information is stored in advance for each behavior modification stage, and further includes an output unit that outputs guidance information according to the behavior modification stage determined by the determination unit.

  Thereby, since it is possible to provide the instructor with optimal guidance information for prompting the subject to change his / her behavior, it is possible to further improve the efficiency and homogeneity of the interactive guidance.

  The present invention may be considered as an interactive instruction support device having at least a part of the above means, or as an interactive instruction support system including a terminal capable of communicating via a computer network and an interactive instruction support server. Good. The present invention can also be understood as an interactive instruction support method including at least a part of the above processing, a program for causing a computer to execute each step of the method, or a recording medium on which the program is recorded. Each of the above means and processes can be combined with each other as much as possible to constitute the present invention.

  For example, in the interactive teaching method according to the present invention, a computer acquires an utterance recording data in which a conversation between a leader and a target person in the interactive teaching is recorded, and a conversation included in the utterance recording data is written. A continuity evaluation step that evaluates the continuity between two sentences and the continuity between the two sentences, and a continuity evaluation step that evaluates the continuity between the two sentences. Based on the gender, a feature value that calculates the degree of influence, which is a feature value that represents the strength of influence on other sentences, and a degree of relationship, which is a feature value that represents the strength of the relationship with other sentences, for each sentence A calculation step, a distribution map creating step of mapping a feature amount of a sentence uttered by the subject to a feature amount space composed of an influence degree and a relevance degree, and a distribution pattern of the feature amount in the feature amount space The basis is for performing a judgment step of judging behavior change stage of the subject.

  According to the present invention, it is possible to automatically determine the behavior change stage without imposing a burden on the subject. As a result, it is possible to improve the efficiency of interactive guidance and to homogenize the guidance content.

  Exemplary embodiments of the present invention will be described in detail below with reference to the drawings. The interactive instruction support apparatus according to the present embodiment is an apparatus that automatically determines an action change stage (consciousness level) of a target person from utterance record data in which a conversation between the instructor and the target person (client) is recorded. For example, it is assumed that the judgment results are output on the spot at the end of the counseling, which can be used for the formulation of instruction content (online analysis), or the counseling records can be analyzed later for the next instruction (offline analysis). ing. This device can be used for various types of interactive guidance, such as health guidance provided by a public health nurse.

<System configuration>
FIG. 1 is a functional block diagram of a health guidance support apparatus according to an embodiment of the present invention.

  The health guidance support device includes an utterance record acquisition unit 10, a decomposition unit 11, a continuity evaluation unit 12, a feature amount calculation unit 13, a distribution map creation unit 14, a determination unit 15, and a guidance information output unit 16. The health guidance support device can be composed of a general-purpose computer including a CPU (central processing unit), a memory (main storage device), an auxiliary storage device, a display device, an input device, and the like. Alternatively, it is realized by executing a program stored in the auxiliary storage device and controlling various hardware resources as necessary.

(Speech record acquisition part)
The utterance record acquisition unit 10 has a function of reading utterance record data into a memory. The utterance record data is read from an auxiliary storage device or an external storage medium, or read from another computer via a computer network.

  FIG. 2A shows an example of utterance record data. The utterance record data includes at least the utterances of the instructor and the target person. In the example of FIG. 2A, the utterance of the instructor starts with a symbol →, and the utterance of the subject is surrounded by the symbol <>, so that both utterances can be easily distinguished. The format of the utterance record data is not limited to this, and may be text data structured with tags or the like, or binary format data. It is also preferable to use voice data obtained by recording a conversation as the utterance recording data.

(Disassembly part)
The disassembling unit 11 has a function of analyzing utterance record data using sentence analysis rules and disassembling conversations included in the utterance record into sentence units. Simply, it is possible to determine the end of a sentence by detecting a specific character such as a line feed character, a punctuation mark, a question mark, an exclamation mark, or a period. When voice data is given as utterance recording data, the decomposition unit 11 converts the voice data into text data by voice recognition, and then performs a decomposition process into sentence units.

  FIG. 2B shows an example of sentence data obtained by disassembling the utterance record data of FIG. 2A. As shown in this example, each sentence data is stored in a state associated with information indicating a speaker.

(Continuity Evaluation Department)
FIG. 3 is a block diagram showing a detailed configuration of the continuity evaluation unit 12. As shown in FIG. 3, the continuity evaluation unit 12 includes (1) a determination processing unit 120 that evaluates continuity between two sentences based on an evaluation rule, and (2) an evaluation rule using learning data. It is comprised from two functions of the determination condition production | generation part 121 which produces | generates automatically.

  Here, first, the continuity evaluation process by the determination processing unit 120 will be described with reference to FIGS. 3 to 6, and the determination condition generation unit 121 will be described later. FIG. 4 is a flowchart showing the flow of continuity evaluation processing, and FIG. 5 is a diagram showing an example of evaluation rules used for continuity evaluation. Moreover, FIG. 6 is a figure which shows an example of a continuity evaluation result.

  The sentence data obtained by the decomposition unit 11 is stored in the sentence 1 storage unit 122 and the sentence 2 storage unit 123, respectively. The determination processing unit 120 reads the sentence 1 and the sentence 2 to be evaluated from the sentence 1 storage unit 122 and the sentence 2 storage unit 123, respectively, and collates with the evaluation rule 124 (step S40).

As shown in FIG. 5, the evaluation rule is data in a format including a determination condition and a set value. In principle, the determination condition defines a combination of a first character string that the sentence 1 should include and a second character string that the sentence 2 should include. When the sentence 1 to be evaluated includes the first character string and the sentence 2 includes the second character string, it is determined that there is continuity between the sentence 1 and the sentence 2. Here, “continuity” is a scale indicating the degree of establishment of two sentences as a conversation. The set value is a value representing the strength of continuity (establishment degree), and takes a value in the range of 0 to 1. A value of 0 indicates a combination that is not clearly established as a conversation, and indicates that continuity is stronger as the value approaches 1.

  For example, in the evaluation rule 1 in FIG. 5, “* Tone * How is it?” Is set as the first character string, and “* Okay *” is set as the second character string. The symbol “*” is a wild card (arbitrary character string). That is, if sentence 1 includes the word “tone” and ends with the description “how?”, And sentence 2 includes the word “OK”, then sentence 1 and sentence 2 are evaluated rules. 1 is satisfied and the continuity is determined to be 1.

  When all the evaluation rules are collated and an evaluation rule that satisfies the determination conditions of sentence 1 and sentence 2 is found (step S41; Yes), the setting value of the corresponding rule is changed from sentence 1 to sentence 2. An evaluation value representing the strength of continuity is set (step S42). On the other hand, if there is no evaluation rule that satisfies the determination conditions for sentence 1 and sentence 2 (step S41; No), the evaluation value from sentence 1 to sentence 2 is set to 0 (step S43).

  The processes in steps S40 to S43 are repeated for all combinations of sentences obtained by the decomposition unit 11. Thereby, continuity is evaluated for all combinations of two sentences, and the result is stored in the evaluation result storage unit 125 of FIG.

  FIG. 6 shows the continuity evaluation results in a matrix. The numerical values in the matrix represent the continuity evaluation value from the vertical nth sentence to the horizontal mth sentence. In this example, there is continuity (evaluation value = 1) from "Vertical 1: 10,000 steps are all right?" To "Horizontal 2: Yes, I walk around a little." Although the evaluation results are obtained, there is no continuity (evaluation value = 0) from "Vertical 1: 10,000 steps are all right?" To "Horizontal 6: Japanese food is fine" You can see that

(Feature amount calculation part)
The feature amount calculation unit 13 is a function that calculates the feature amount of each sentence based on the continuity obtained by the continuity evaluation unit 12. Here, as the feature amount of a sentence, “impact level” that is a feature amount indicating the strength of influence on other sentences and “relevance” that is a feature amount indicating the strength of the relation with other sentences are described. Two are used.

  Here, with reference to FIG. 7, a method of calculating the influence degree and the relation degree will be described. 7 shows an example of calculation with five sentences for the sake of simplicity of explanation, but in the case of actual utterance record data including tens to hundreds of sentences, the influence and relevance are calculated in the same manner. Is possible.

  FIG. 7A is a matrix of continuity evaluation values obtained by the continuity evaluation unit 12. In this example, evaluation values of five combinations of “sentence 1 → sentence 2”, “sentence 2 → sentence 3”, “sentence 2 → sentence 4”, “sentence 3 → sentence 4”, and “sentence 5 → sentence 4”. Is 1. First, the feature quantity calculation unit 13 evaluates the reachability between sentences based on this matrix, and generates the reachability matrix shown in FIG. 7B. Reachability means that sentences have continuity directly or indirectly. For example, “sentence 1 → sentence 1” between the same sentences and “sentence 1 → sentence 2” having direct continuity are reachable combinations. Further, by sequentially tracing “Sentence 1 → Sentence 2” and “Sentence 2 → Sentence 3”, the sentence 1 is connected to the sentence 3, so that “Sentence 1 → Sentence 3” is also a reachable combination. In the reachable matrix of FIG. 7B, the value of the reachable combination is set to 1, and the value of the unreachable combination is set to 0.

  Next, the feature quantity calculation unit 13 adds the reachability for each row of the reachability matrix and obtains the D value of each sentence. In addition, the reachability is added up for each column, and the R value of each sentence is obtained. Subsequently, the feature amount calculation unit 13 calculates a value of (D + R) and a value of (D−R) for each sentence. The value of (D + R) is the degree of association, and the value of (D−R) is the degree of influence.

(Distribution map creation part)
The distribution map creation unit 14 has a function of creating a distribution map by mapping feature quantities of a sentence uttered by a target person in a feature quantity space composed of an influence degree and a relation degree.

  FIG. 8 shows an example of the distribution map. The horizontal axis is the degree of association, and the vertical axis is the degree of influence. The value of influence and relevance shows a certain tendency according to the content of the sentence. For example, an essential question (Know-Why) sentence such as “What should be removed from obesity?” Tends to have a high degree of influence and a low degree of relevance. The sentence describing a specific method / issue (Know-How) such as “.” Has a tendency that the degree of influence takes a negative value. Sentences that describe impressions tend to have a greater influence and relevance. Therefore, the tendency of the content of the subject's utterance appears in the feature quantity distribution in this distribution map.

(Judgment part)
The determination unit 15 has a function of determining the behavior change stage (consciousness level) of the target person based on the feature amount distribution pattern in the feature amount space. Specifically, the determination unit 15 includes a template that defines a distribution pattern corresponding to each behavior change stage in advance, and matches the distribution map obtained by the distribution map creation unit 14 with each template. Identify the behavior change stage to be performed.

  FIG. 9 shows an example of a template. The template of the “indifference period” stage is a pattern in which feature amounts are concentrated in a region where the influence degree is negative. In the indifference period, the content of the story is biased toward specific or simple answers, and the speaker himself rarely asks questions, so this distribution pattern often occurs. The template of the “interest period” stage is a pattern in which feature quantities appear in a dispersed manner. This means that the speaker's interests and topics have become diverse. The template in the “preparation stage” is a pattern in which the feature amount is concentrated in a region where the influence degree is positive. In the preparatory stage, there are many questions from the speakers and the provision of topics. On the other hand, there are few specific stories because no action has been started, so this pattern is used. The template for the “behavioral period” stage is a pattern in which feature amounts are concentrated in two places, a positive influence area and a negative influence area. Such a balanced pattern appears because the conversation of the speaker who is actually acting appropriately includes interests and questions about the guidance content and specific topics such as reporting of the behavior content.

  When collating with the template, the determination unit 15 first clusters the feature amount distribution of the target person. When the number of clusters is 1 and a cluster exists in an area where the influence degree is negative, it is determined as “indifference period”. If the number of clusters is 1 and there is a cluster in a positive impact area, it is determined as “preparation period”. If the number of clusters is 2 and each cluster is moderately separated into positive and negative influence levels, it is determined as “action period”. When the number of clusters is 2, but the distance between the clusters is small, it is classified as “preparation period” or “indifference period”. When the number of clusters is 3 or more, it is determined as “interesting period”.

  The determination result of the determination unit 15 is output to the display device. By looking at the determination result, the instructor can grasp the behavior change stage (consciousness level) of the target person, and can use it for the formulation of instruction content. Furthermore, the determination result of the determination unit 15 is also delivered to the guidance information output unit 16.

(Guidance information output part)
The guidance information output unit 16 is a function that outputs optimal guidance information according to the behavior change stage determined by the determination unit 15. As shown in FIG. 1, this apparatus includes an assist rule base that stores in advance guidance information created based on past cases and know-how. Guidance information includes information, materials, manuals, and the like for assisting interactive guidance by a leader (health nurse). In this example, in the assist rule base, data with the name “fear of obesity” for those who are not interested, and data “recommendation of exercise” for those who are interested, Data for "planning guide" is prepared for the target, and data for "comment manual" is prepared for the target person in the action period. As long as browsing is possible on the apparatus, any data format may be used for the guidance information. For example, HTML, XML, PDF, plain text may be used, and data such as words, Excel, and PowerPoint may be used.

  By viewing the output guidance information, the leader can provide guidance suitable for the behavior change stage of the subject.

<Generation of evaluation rules>
Next, an evaluation rule generation process by the determination condition generation unit 121 of the continuity evaluation unit 12 will be described in detail.

  FIG. 10 is a block diagram illustrating a detailed configuration of the determination condition generation unit 121. The determination condition generation unit 121 includes a sentence type identification unit 1210, a similar sentence extraction unit 1211, and a condition generation unit 1212. The determination condition generation unit 121 refers to learning data, a sentence type identification table, a thesaurus, and a collection of prohibited words. Generate.

  FIG. 11 is a diagram for explaining processing of the sentence type identification unit 1210. As the learning data, a set of sentence pairs (sentence 1, sentence 2) having continuity is given. Typically, conversational sentences extracted from past utterance records of health guidance are used as learning data. In order to ensure the reliability of the evaluation rule, it is preferable to prepare as much learning data as possible.

  The sentence type identification unit 1210 identifies the type of each sentence included in the learning data. The types of sentences include “question”, “response”, “greeting”, “praise”, “suggestion”, “agreement”, and the like. In the sentence type identification table, a table in which sentence types and their representative expressions are associated with each other is prepared in advance, and the sentence type identification unit 1210 collates descriptions such as sentence endings with respective representative expressions. To identify the type of sentence. For example, no. Sentence 1 “How is it?” Is identified as “Question” based on the description “How is it?”, And Sentence 2 “I think it ’s messy” is “I think.” Is identified as a “response”. This No. Information representing the combination “question → answer” of the type of sentence 1 and the type of sentence 2 is assigned to the pair of 1. Similarly, sentence types are assigned to all pairs included in the learning data. The underline in the sentence in the lower part of FIG. 11 indicates the description that is the basis for determining the sentence type.

FIG. 12 is a diagram for explaining the processing of the similar sentence extraction unit 1211. The similar sentence extraction unit 1211 extracts pairs having the same combination of sentence types from the learning data. The upper part of FIG. 12 is an example of extracting a pair of “question → response”. Next, the similar sentence extraction unit 1211 refers to the thesaurus, and extracts pairs in which the same or similar words are included in each of the sentences 1 and 2. For example, no. When the sentence 1 “How is the tone?” And the sentence 2 “I think it's a little bit” are used as a reference, the words “Tone” from the sentence 1 and “Bochobichi” from the sentence 2 are extracted. Then, a pair that includes the same word or synonym as “tone” in sentence 1 and includes the same word or synonym as “bochobuchi” in sentence 2 is extracted as a similar sentence. In the example of FIG. 12, “tone” and “anbai”, “botchiboko”, “so-so”, and “decent” are determined as synonyms from the thesaurus. Three pairs of 1, 2, and 5 are extracted. Note that the rectangles in the lower sentence of FIG. 12 indicate words that are the basis for determining similar sentences.

  FIG. 13 is a diagram for explaining the process of the condition generation unit 1212. The condition generation unit 1212 generates a determination condition from each pair extracted by the similar sentence extraction unit 1211. At this time, a determination condition is generated from the word selected at the time of similar sentence extraction and the description taken into consideration at the time of sentence type determination. In the sentence type identification table, a table in which “sentence type” is associated with “information indicating whether or not a description is reflected in a condition” is prepared in advance. For the type of sentence that is likely to be omitted, such as “Reply” and “Agree”, “No” is set in the information indicating whether the description is reflected in the condition.

  For example, no. In the case of one pair, the word “tone” and the description “how?” Are extracted from sentence 1, and the first character string “* tone * how?”, Which is a condition for sentence 1, is generated. On the other hand, only the word “bottom” is extracted from sentence 2 (because the type of sentence 2 is a response), and the second character string “* botchi *” that is a condition for sentence 2 is generated. As a result, the judgment condition “S1 =“ * Tone * How is it? An evaluation rule of “AND S2 =“ * Bobchi * ”” is generated.

  Through the above processing, after the evaluation rules are automatically generated from the learning data, the user adds / edits the evaluation rules as necessary. FIG. 14 shows an example of the created evaluation rule. Of these, RuleNo. 1, 2, 5, and 7 are rules automatically generated by the above-described processing. 3, 4, 6, 8, 9, and 10 are manually added rules. In FIG. 14, “S1” and “S2” are sentence 1 and sentence 2 to be evaluated, respectively. “P1 AND p2” indicates that if both p1 and p2 are true, it is true. “S =“ str ”” indicates that the sentence s is true if it is equal to the character string str (* is a wild card). “S.preSensence.valueWith (s ′)” is a setting value between the sentence s ′ and the sentence immediately preceding the sentence s. “S.words” is a set of words included in the sentence s. “S.words.containAnyof (words)” is true if the set of words included in the sentence s includes any of the words that are not prohibited words of the word set words (or its synonyms, broad / narrow sense terms, related terms). Indicates that “S.words.contain (word)” indicates that it is true if the set of words included in the sentence s includes the word word (or its synonyms, broader / narrower terms, related terms). FIG. 15 shows an example of a thesaurus and an example of a prohibited word collection.

  The evaluation rule created in this way and the evaluation method using the rule are particularly suitable for evaluating the continuity of a conversation sentence. Conversational sentences are abbreviated and anaphoric, and contain interjections and inversions. Therefore, only general morphological analysis (words) and syntactic analysis (such as dependency) require context (ie, sentence continuity). It is difficult to evaluate. On the other hand, the evaluation rule and the evaluation method of this embodiment are (1) the type of sentence can be identified from the ending of the conversation sentence, and (2) the combination of the types of sentences that can be continued is determined. (3) If two sentences are in a relationship of combinations of sentence types that can be continued, and there is a relationship between a word contained in one sentence and a word contained in the other sentence, the two sentences are consecutive. It is based on the idea that it can be regarded as such.

In other words, the process in which the similar sentence extraction unit 1211 extracts pairs having the same combination of sentence types from the learning data is equivalent to detecting “a combination of consecutive sentence types” in (2). . Moreover, the process in which the similar sentence extraction part 1211 extracts the pair in which the same word or synonym is included detects the combination of the word (a synonym is included) with high frequency which co-occurs in the sentence 1 and the sentence 2. Is equivalent. The high frequency of co-occurring in two sentences can be considered to be related to the two words. Therefore, the process of evaluating using the determination condition including the word selected at the time of similar sentence extraction and the description taken into account at the time of sentence type determination is the above-mentioned (3) “Combination of sentence types that two sentences can be continuous. This is equivalent to the process of evaluating whether or not there is a relationship between a word included in one sentence and a word included in the other sentence.

<Example>
Then, the advantage of this apparatus is demonstrated using a specific example.

  For each of the following two utterance record examples, the behavior change stage of the subject is determined. Note that the speech contents of the public health nurse are the same in both Example 1 and Example 2, and the number of sentences spoken by the subject is the same. At first glance, it is difficult to determine where the subject's behavior change stage is.

(Example 1)
Public health nurse: How are you feeling recently?
Target audience: OK.
Public health nurse: Do you have anything ingenuity?
Target: I'm taking a walk on holidays. Besides, I eat vermicelli noodles every day. I went to the gym, how are you? He also drinks almost every day.
Public health nurse: However, if you look at the records for the past month, your weight seems to be increasing.
Subject: Well, I wonder if there are not enough steps. Or is it because there are many fried foods in the daytime?
Public health nurse: What are your thoughts as a countermeasure?
Subject: I feel that swimming is more likely to work than walking, recently. I will also try the carbohydrate diet I asked my wife.

(Example 2)
Public health nurse: How are you feeling recently?
Target audience: OK.
Public health nurse: Do you have anything ingenuity?
Subject: As I was told the other day, I'm already old and my autonomic nerves are weakened. First of all, I want to strengthen my autonomic nerves. Specifically, take a walk on holidays. Use stairs to move at work on weekdays.
Public health nurse: However, if you look at the record for the past month, your weight seems to be increasing.
Target audience: Well, I don't think there are any particular restrictions on food. There are many oily fried foods in the daytime.
Public health nurse: What are your thoughts as a countermeasure?
Audience: Let's refrain from fried food. First, try not to eat your favorite tonkatsu.

(1) Case of Example 1 FIG. 16 shows a result of evaluating the continuity of the sentence of Example 1 using the evaluation rule of FIG. 14 and the thesaurus and prohibited word collection of FIG. For example, “How are you feeling recently?” And “Okay”, RuleNo. 1 is satisfied. Also, “Are there anything ingenuity?” And “I take a walk on holidays.” 9 is satisfied (“walk” is a related word for “devise”). FIG. 17 shows a distribution map created based on the evaluation result of FIG. In this distribution map, since the feature amount is concentrated in the area where the influence degree is negative, it is determined as “indifference period” in light of the template of FIG. 9.

(2) Case of Example 2 FIG. 18 shows a result of evaluating the continuity of the sentence of Example 2 using the evaluation rule of FIG. 14 and the thesaurus and prohibited word collection of FIG. A distribution map created based on this is shown in FIG. In this distribution map, since the distribution of the feature amount appears in each of the positive influence area and the negative influence area, it is determined as the “action period”.

<Advantages of this embodiment>
According to the configuration of the present embodiment described above, the behavior change stage (consciousness level) of the target person is automatically determined from the utterance record data that records the conversation between the instructor (specialist such as a public health nurse) and the target person. can do. Therefore, the instructor can efficiently concentrate on the conversation with the target person during the face-to-face instruction. In addition, since the behavior change stage of the target person is automatically and objectively determined, it is possible to make the content of the guidance uniform regardless of the experience and skills of the leader.

  Further, according to the evaluation rule, it is possible to evaluate the continuity between sentences by a simple process. Furthermore, since the continuity is evaluated by paying attention to the relationship between the type of sentence and the word, the continuity of the conversation sentence can be accurately determined. This is extremely useful for analysis of utterance record data in interactive guidance.

  In addition, by outputting appropriate guidance information according to the determined behavior change stage, it is possible to provide the leader with optimal guidance information for encouraging the subject to change their behavior. And homogenization can be achieved.

  The above embodiment is merely a specific example of the present invention. The scope of the present invention is not limited to the above embodiment, and various modifications are possible.

FIG. 1 is a functional block diagram of a health guidance support apparatus according to an embodiment of the present invention. FIG. 2A is an example of utterance record data, and FIG. 2B is an example of sentence data obtained by disassembling the utterance record data. FIG. 3 is a block diagram illustrating a detailed configuration of the continuity evaluation unit. FIG. 4 is a flowchart showing the flow of continuity evaluation processing. FIG. 5 is a diagram illustrating an example of evaluation rules used for continuity evaluation. FIG. 6 is a diagram illustrating an example of a continuity evaluation result. FIG. 7 is a diagram for explaining a method for calculating the degree of influence and the degree of association. FIG. 8 is a diagram illustrating an example of a distribution map. FIG. 9 is a diagram illustrating an example of a template. FIG. 10 is a block diagram illustrating a detailed configuration of the determination condition generation unit. FIG. 11 is a diagram for explaining processing of the sentence type identification unit. FIG. 12 is a diagram for explaining processing of the similar sentence extraction unit. FIG. 13 is a diagram for explaining the processing of the condition generation unit. FIG. 14 is a diagram illustrating an example of an evaluation rule. FIG. 15 is a diagram illustrating an example of a thesaurus and an example of a forbidden word collection. FIG. 16 shows the continuity evaluation result of the sentence in Example 1. FIG. 17 is a sentence distribution map of Example 1. FIG. 18 shows the continuity evaluation result of the sentence in Example 2. FIG. 19 is a sentence distribution map of Example 2.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Utterance record acquisition part 11 Decomposition part 12 Continuity evaluation part 13 Feature-value calculation part 14 Distribution map creation part 15 Determination part 16 Guidance information output part 120 Determination processing part 121 Determination condition generation part 122 sentence 1 memory | storage part 123 sentence 2 memory | storage part 124 Evaluation rule 125 Evaluation result storage unit 1210 Sentence type identification unit 1211 Similar sentence extraction unit 1212 Condition generation unit

Claims (7)

An utterance record acquisition unit for acquiring utterance record data in which conversation between a leader and a subject in interactive guidance is recorded;
A disassembling unit for disassembling a conversation included in the utterance record data into sentence units;
A continuity evaluation unit that evaluates continuity between two sentences for all combinations of sentences obtained by the decomposition unit;
Based on the continuity obtained by the continuity evaluation unit, the degree of influence that is a characteristic amount that represents the strength of influence on other sentences and the degree of association that is a characteristic quantity that represents the strength of association with other sentences A feature amount calculation unit for calculating
A distribution map creation unit that maps the feature amount of the sentence spoken by the target person to the feature amount space composed of the influence degree and the relevance degree;
A determination unit that determines a behavior modification stage of the target person based on a distribution pattern of the feature amount in the feature amount space;
An interactive teaching support device characterized by comprising:   The continuity evaluation unit has in advance an evaluation rule including a determination condition that defines a combination of a first character string and a second character string, and the first sentence of two sentences is the first sentence. 2, and when the second sentence includes the second character string, it is determined that the two sentences have continuity. The interactive teaching support device described. A determination condition generation unit that automatically generates the evaluation rule determination condition using learning data;
The learning data is composed of a pair of a first sentence and a second sentence having continuity,
The determination condition generation unit
Identifying the type of each sentence included in the learning data based on the description included in the sentence;
Among the learning data, the combination of the first sentence type and the second sentence type is the same, and each of the first sentence and the second sentence includes the same or similar word. Extract the pair that is
From each of the extracted pairs, a determination condition is generated in which the description or / and the word in the first sentence is the first character string, and the description or / and the word in the second sentence is the second character string. The interactive instruction support apparatus according to claim 2, wherein:   The determination unit includes a template that defines a distribution pattern corresponding to each behavior modification stage in advance, and identifies a corresponding behavior modification stage by comparing the distribution pattern of the feature amount in the feature amount space with each template. The interactive instruction support apparatus according to any one of claims 1 to 3, wherein   The guidance information is previously memorize | stored for every action change stage, The output part which outputs guidance information according to the action change stage determined by the said determination part is further provided, The any one of Claims 1-4 characterized by the above-mentioned. The interactive instruction support device according to claim 1. Computer
An acquisition step of acquiring utterance record data in which conversation between the leader and the subject in interactive guidance is recorded;
A disassembling step of disassembling a conversation included in the utterance record data into sentence units;
A continuity evaluation step for evaluating continuity between two sentences for all combinations of sentences obtained in the decomposition step;
Based on the continuity obtained in the continuity evaluation step, the degree of influence, which is a feature amount representing the strength of influence on other sentences, and the degree of association, which is a feature quantity representing the strength of association with other sentences, A feature amount calculating step for calculating each sentence,
A distribution map creating step for mapping a feature amount of a sentence uttered by the target person to a feature amount space composed of an influence degree and a relevance degree;
A determination step of determining a behavior modification stage of the subject based on a distribution pattern of the feature amount in the feature amount space;
An interactive teaching support method characterized by executing   A program for causing a computer to execute each step of the interactive instruction support method according to claim 6.

Images