Conversation evaluation device

Abstract

To correctly evaluate a risk in conversation.SOLUTION: A conversation evaluation device according to the present disclosure includes: an intention database in which previously set intention phrases being phrases of intention expression by speakers are registered; an exclusion database in which previously set excluded phrases being phrases containing at least a part of a declarative sentence at least partly containing an intention phrase are registered; a specifying unit which extracts a character string part containing the intention phrase from acquired conversation data on the basis of the intention database, and specifies an objective character string obtained by excluding a character string part containing the excluded phrases from the extracted character string part on the basis of the exclusion database; and an evaluation unit which evaluates the conversation data on the basis of the objective character string.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a conversation evaluation device, a conversation evaluation method, and a program for evaluating the risk level of a conversation.

Background Art

[0002] In recent years, crimes such as fraud through telephone calls have been increasing. Examples of such crimes include so-called remittance fraud, where a criminal pretends to be a family member of the call recipient and deceives them into making a money transfer. In the case of such remittance fraud, the criminal often uses specific phrases such as "I" (without stating their name) and phrases related to "money" or "transfer" because they are pretending to be a family member and requesting a money transfer. In addition, calls during other criminal acts such as fraud often contain phrases specific to the content of the crime.

[0003] Therefore, in order to suppress crimes such as fraud through telephone calls as described above, as described in Patent Document 1, speech recognition of conversation audio is performed to extract preset keywords. Patent Document 1 describes evaluating the suspiciousness of a call based on the number of occurrences of the extracted keywords and notifying the user.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, in the above-described technology, since the suspiciousness is evaluated based on the extracted keywords, if the call does not contain any keywords, the suspiciousness will be evaluated as low. As a result, there is a problem that risks such as the suspiciousness of the call cannot be accurately evaluated. Further, not only in the case of a call by telephone, but also in any conversation such as a conversation via an intercom or a conversation through a door, the above-described problem occurs.

[0006] Therefore, an object of the present invention is to provide a conversation evaluation apparatus that can solve the above-described problem of being unable to accurately evaluate the risk of a conversation.

Means for Solving the Problem

[0007] A conversation evaluation apparatus according to one embodiment of the present invention includes an intention database in which preset intention phrases, which are phrases for expressing intentions by a speaker, are registered, an exclusion database in which preset exclusion phrases, which are phrases consisting of at least a part of a declarative sentence including at least a part of the intention phrase, are registered, a specifying unit that extracts a character string portion including the intention phrase based on the intention database from the acquired conversation data, and specifies a target character string obtained by excluding the character string portion including the exclusion phrase based on the exclusion database from the extracted character string portions, an evaluation unit that evaluates the conversation data based on the target character string, and is configured to take the above configuration. Further, a conversation evaluation method according to one embodiment of the present invention is an information processing apparatus including an intention database in which preset intention phrases, which are phrases for expressing intentions by a speaker, are registered, and an exclusion database in which preset exclusion phrases, which are phrases consisting of at least a part of a declarative sentence including at least a part of the intention phrase, are registered, is configured to Extract, from the obtained conversation data, string locations that include the intention phrase based on the intention database, and identify a target string obtained by excluding, from the extracted string locations, string locations that include the exclusion phrase based on the exclusion database. Evaluate the conversation data based on the target string. It has such a configuration. Also, a program according to an aspect of the present invention An intention database in which preset intention phrases that are phrases for expressing intentions by a speaker are registered, An exclusion database in which preset exclusion phrases that are phrases consisting of at least a part of declarative sentences including at least a part of the intention phrase are registered, In an information processing apparatus provided with Extract, from the obtained conversation data, string locations that include the intention phrase based on the intention database, and identify a target string obtained by excluding, from the extracted string locations, string locations that include the exclusion phrase based on the exclusion database. Evaluate the conversation data based on the target string. Cause to execute the process. It has such a configuration.

Advantages of the Invention

[0008] With the present disclosure configured as described above, it is possible to accurately evaluate the risk of a conversation.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Mode for Carrying Out the Invention

[0010] <Embodiment 1> The first embodiment of the present disclosure will be described with reference to the drawings. Note that the drawings are merely examples, and the present disclosure is not limited to the content described in the drawings.

[0011] The call evaluation device (conversation evaluation device) in the present disclosure calculates a risk level representing the degree of risk (degree of danger) of a call (conversation) from the content of a call (conversation) using a telephone, and is for evaluating whether the call is a high-risk call. In particular, in this embodiment, for example, it evaluates the possibility that a call using a telephone is a so-called transfer fraud call where a criminal pretends to be a family member of the call partner and requests a money transfer, or the possibility that the call is a criminal call for stealing financial account information. However, the call evaluation device according to the present invention may evaluate the possibility of any risky call such as abduction or illegal business. In this embodiment, a call using a telephone is taken as an example for the object of evaluating the degree of risk, but in the present disclosure, it may be used to evaluate the risk of any conversation such as a conversation via an intercom or a conversation through a door.

[0012] Here, the call evaluation device in the present embodiment described below evaluates whether a call received by the user terminal 1, which is the mobile phone of the user P shown in FIG. 1, is a suspicious call. For this reason, the call evaluation device is constituted by the user terminal 1 itself, and specifically, it is realized by a program incorporated in the user terminal 1.

[0013] The user terminal 1 is an information processing device including an arithmetic unit and a storage unit, such as a smartphone operated by user P. Then, the user terminal 1 downloads and installs an application program for call evaluation provided by a predetermined operator from a predetermined web server, and as shown in FIG. 1, includes an acquisition unit 11, an analysis unit 12, and an evaluation unit 13. That is, the acquisition unit 11, the analysis unit 12, and the evaluation unit 13 are realized by the arithmetic unit executing the application program. Further, when the application program is executed, the user terminal 1 stores various databases included in the application program, and a phrase storage unit 16 is configured in the storage unit. Furthermore, when the application program is executed, the user terminal 1 includes a call data storage unit 17 configured in the storage unit. Hereinafter, each configuration and operation will be described.

[0014] First, when the above-described application program for call evaluation is installed, the user terminal 1 stores an intention database, an exclusion database, and a noun database in the phrase storage unit 16 (step S1 in FIG. 5).

[0015] The intention database stores preset intention phrases that are phrases of intention expressions by the caller. Here, FIG. 2 shows an example of the intention phrases registered in the intention database. The intention phrases in the present embodiment are, in particular, phrases of intention expressions classified into the contents of "confirmation", "instruction", and "request" by the caller to the other party of the call. As an example, the intention phrases of the "confirmation" intention expression are "desuka", "aruno", etc., the intention phrases of the "instruction" intention expression are "naide", "tene", etc., and the intention phrases of the "request" intention expression are "kudasai", "moraeru", etc. Note that the intention phrases are not limited to the phrases of intention expressions classified into the above-described "confirmation", etc., and may be any phrases representing the intention of the caller to the other party of the call.

[0016] Note that the intent phrase is generated and registered in the intent database as follows. First, set the endings where an intent (such as confirmation, instruction, or request) is assumed to appear. For example, set endings such as "··· (desu) ka", "··· (na) no" for the intent of confirmation, "··· (shi) te", "··· (shi nai) de" for the intent of instruction, and "··· (shi te) kudasai", "··· (shi te) mora itai" for the intent of request. Subsequently, set the part-of-speech conditions for the cases where each ending indicates an intent. As an example, when the set ending is "ka", set part-of-speech conditions such as "Extract the single 'ka' from the morphological analysis result, and if its part of speech is a particle (adverbial particle / coordinating particle / terminal particle) and the part of speech before 'ka' is a verb, auxiliary verb, or adjective, mark 'ka'". As another example, when the set ending is "no", set part-of-speech conditions such as "Extract the single 'no' from the morphological analysis result, and if its part of speech is a particle (terminal particle), mark 'no', or if its part of speech is not a particle (terminal particle) but a particle (attributive) or a noun, mark 'no' if the part of speech immediately before 'no' is a verb, auxiliary verb, or adjective".

[0017] Then, from the call text data obtained by converting the sample call data into text by voice recognition, extract and register the intent phrase according to the ending that meets the above-mentioned part-of-speech conditions. Specifically, first, extract the ending that meets the part-of-speech conditions from the sample call text data, and also extract the two parts of speech before and after such an ending. For example, from call data such as "··· at home desu ka watakushi city hall ···", extract the ending "ka" that meets the quality conditions, and extract the two parts of speech "at home", "desu" before it and the two parts of speech "watakushi", "city hall" after it. Then, extract and register the phrase where the intent appears as the intent phrase from the extracted phrases. In this case, register "desu ka" as the intent phrase. The intent database in which the intent phrases as shown in Figure 2 generated in this way are registered is stored in the user terminal 1.

[0018] The exclusion database stores exclusion phrases generated based on the above-described intended phrases. Here, FIG. 3 shows an example of an exclusion phrase registered in the exclusion database. The exclusion phrases in this embodiment are generated from declarative sentences that include at least a part of the intended phrases, and are phrases that include at least a part of such declarative sentences. In particular, the exclusion phrase may be a phrase that includes the end of a declarative sentence consisting of a character string in which other characters are concatenated to the end of the intended phrase, and may further be a phrase that includes the end of the intended phrase. However, the exclusion phrase is not limited to those shown in FIG. 3, and any phrase may be used as long as it is generated from a declarative sentence that includes a part of the intended phrase.

[0019] Note that the exclusion phrases are generated and registered in the exclusion database as follows. First, the intended phrases are extracted from the call text data obtained by textifying sample call data, and further, declarative sentences that include a part of such intended phrases are extracted. For example, when the intended phrase "aruno" (confirmation) is extracted, if the character string "arunode" in which other characters are concatenated to the end of such intended phrase is a declarative sentence, such declarative sentence is extracted. Then, the end "node" of the declarative sentence "arunode" that has been extracted is set as the exclusion phrase. As another example, when the intended phrase "naide" (instruction) is extracted, if the character string "naidesu" in which other characters are concatenated to the end of such intended phrase is a declarative sentence, such declarative sentence is extracted. Then, the end "desu" of the declarative sentence "naidesu" that has been extracted is set as the exclusion phrase. The exclusion database in which the exclusion phrases as shown in FIG. 3 generated in this way are registered is stored in the user terminal 1.

[0020] The noun database stores phrases consisting of preset nouns. The registered nouns are phrases that have been pre-determined to potentially be used in crimes such as the above-described transfer fraud. As an example, "ore", "money", "transfer", etc. are included. However, the noun database may store phrases consisting of any nouns.

[0021] Then, the acquisition unit 11 of the user terminal 1 acquires call data (conversation data), which is the voice data of a call (conversation) being made on the user terminal 1. At this time, the acquisition unit 11 performs speech recognition on the call data, which is the voice data acquired by the speech recognition function, to convert it into text, and stores the call text data together with the call data in the call data storage unit 17 (step S2 in FIG. 5). However, the acquisition unit 11 may store only the call text data without storing the call data consisting of voice data. Further, the acquisition unit 11 may acquire and store the call text data that has been converted into text by another information processing device connected to the network without converting the call data consisting of voice data into text.

[0022] Subsequently, the analysis unit 12 (specification unit) of the user terminal 1 extracts the intention phrases registered in the intention database from the call text data (step S3 in FIG. 5). Here, FIG. 4 shows an example of the call text data. This text data is a part of an actual fraud call published on the homepage of the Chiba Prefectural Police (https: / / www.police.pref.chiba.jp / index.html). The upper figure in FIG. 4 shows an example of extracting intention phrases from the call text data. In the example shown in the upper figure of FIG. 4, six intention phrases such as "deshouka" (1) (confirmation), "kudasaru" (request) (2), "masuka" (3) (confirmation), "nano" (4) (confirmation), "desuka" (5) (confirmation), and "desuyone" (6) (confirmation) are extracted. Note that the numbers in parentheses ((1), etc.) and the underlines in FIG. 4 are added for the convenience of explanation for the illustration of the extracted intention phrases and have nothing to do with the call.

[0023] Subsequently, the analysis unit 12 excludes those including exclusion phrases from the intention phrases extracted as described above, and identifies the remaining intention phrases that are not excluded as evaluation target phrases (target character strings) (step S4 in FIG. 5). Specifically, the analysis unit 12 first extracts string locations that include the extracted intention phrases and further add characters in the call text data before and after the intention phrases. For example, in the example of FIG. 4, for the intention phrase "nano" (4) extracted in the upper figure, the string location such as "nanode" (4)' is extracted by expanding the end word by one character in the call text data as shown in the lower figure. Also, for the intention phrase "desuka" (5) extracted in the upper figure, the string location such as "desukara" (5)' is extracted by expanding the end word by one character in the call text data as shown in the lower figure. Note that the number of characters expanded from the intention phrase is not limited to one character, and can be any number of characters. Also, the process of extracting the string locations obtained by expanding the character strings from the intention phrases may, for example, expand characters by a preset number of characters, or set and extract the string locations to be expanded from the parts of speech before and after by performing morphological analysis on the call text data.

[0024] Then, the analysis unit 12 checks whether there is any string location obtained by expanding from the intention phrase that includes an exclusion phrase registered in the exclusion database, and excludes those that include an exclusion phrase from the string locations obtained by expanding from the intention phrase. For example, in the example of the lower figure in FIG. 4, the string location "nanode" (4)' is excluded because it includes the exclusion phrase "node", and the string location "desukara" (5)' is excluded because it includes the exclusion phrase "kara". As a result, the string locations "deshouka" (1), "kudasaru" (2), "masuka" (3), and "desuyone" (6) that include intention phrases that are not excluded are identified as evaluation target phrases.

[0025] Also, the analysis unit 12 extracts nouns registered in the noun database from the call text data (step S5 in FIG. 5). Note that such extraction of nouns may be performed in any order.

[0026] Then, the evaluation unit 13 of the user terminal 1 evaluates the risk of the call data based on the evaluation target phrase (target character string) specified as described above and the extracted noun (step S6 in FIG. 5). Specifically, the evaluation unit 13 evaluates the risk of the call according to the appearance status of the specified evaluation target phrase and the extracted noun in the call data. For example, the evaluation unit 13 evaluates that the higher the number of appearances of the evaluation target phrase in the call data, the higher the risk, or the higher the number of appearances of the extracted noun, the higher the risk. In addition, the evaluation unit 13 may evaluate the risk according to the content of the specified evaluation target phrase and the extracted noun. For example, weights are set for each content of the evaluation target phrase and the noun, and the greater the risk of the call data is evaluated as the greater the weight of the evaluation target phrase and the noun, and further, the greater the number of appearances of the evaluation target phrase and the noun with the greater weight.

[0027] However, the method for evaluating the risk of call data by the evaluation unit 13 described above is an example, and any method may be used for evaluation as long as it is an evaluation based on the specified evaluation target phrase and the extracted noun. Also, in the above, the risk of call data is evaluated based on the specified evaluation target phrase and the extracted noun, but the risk evaluation may be performed using only the specified evaluation target phrase without using the extracted noun. That is, the evaluation unit 13 may evaluate the risk according to the appearance status (number of appearances and weight of content) of the specified evaluation target phrase in the call data. In this case, the above-described analysis unit 12 does not have to extract nouns from the call text data.

[0028] Then, the evaluation unit 13 performs processing according to the result of the evaluation performed as described above. For example, the evaluation unit 13 may output to display a numerical value or a diagram representing the degree of risk evaluated on the display device of the user terminal 1, or may output a notification sound notifying that the risk is high (step S7 in FIG. 5). Further, the evaluation unit 13 may output to notify the evaluation result of the risk to an information processing device of another user registered in advance, for example, a family member of user P. Furthermore, when the evaluation unit 13 is performing risk evaluation by acquiring call data as described above while the user is making a call using the user terminal 1, if it is determined that the risk is high such as the risk being equal to or higher than the threshold value, the evaluation unit 13 may perform processing such as disconnecting the call.

[0029] Here, in the above, the case where the call evaluation device is the user terminal 1 through which the user makes a call has been exemplified, but the call evaluation device may be configured by an information processing device different from the terminal through which the user makes a call. For example, when the user makes a call using a landline phone, a transfer device that collects and transfers voice data from the call to the landline phone is attached, and an information processing device such as a smartphone that acquires the call data, which is the voice data transferred from such a transfer device, may configure the call evaluation device. In this case, the transfer device is mounted on the speaker part of the receiver of the landline phone, and includes a microphone that collects the call from the caller who has called the user, and a transfer unit that transfers the collected voice data to an information processing device such as a smartphone by short-range wireless communication such as Bluetooth (registered trademark), and can be realized by this. Note that the above-described transfer device may be, for example, built in the landline phone in advance, and the voice data of the call made by the landline phone may be transferred to the information processing device that is the call evaluation device by any method.

[0030] In addition, in the present disclosure, it is applicable not only to calls using a telephone but also to any conversation. In this case, it can be applied by equipping the portable information processing terminal possessed by the user with the functions of the above-described call evaluation device. As an example, when the user is having a conversation with a conversant through an intercom or through a door, the microphone equipped in the portable information processing terminal picks up the conversation with the conversant to obtain conversation data, and by evaluating such conversation data in the same manner as described above, the risk of the conversation can be evaluated. In this case, the portable information processing terminal may be configured to output a notification sound when it is determined that the risk of the conversation is high.

[0031] With the present disclosure configured as described above, it extracts the intention phrases of the intention expressions by the conversant, excludes those that are not the desired intention with exclusion phrases, and evaluates the risk of the conversation using the remaining phrases. Therefore, the risk of the conversation can be evaluated more accurately.

[0032] As described above, the present disclosure has been described with reference to the above-described embodiments and the like, but the present disclosure is not limited to the above-described embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. And each of the above-described embodiments can be combined with other embodiments as appropriate.

[0033] <Supplementary Note> Some or all of the above-described embodiments can also be described as follows. Hereinafter, an outline of the configuration of the conversation evaluation device, conversation evaluation method, and program in the present disclosure will be described. However, the present disclosure is not limited to the following configuration. (Supplementary Note 1) An intention database in which preset intention phrases that are phrases of intention expressions by a conversant are registered, An exclusion database in which preset exclusion phrases that are phrases consisting of at least a part of a declarative sentence including at least a part of the intention phrase are registered, Extract a string portion containing the intention phrase from the obtained conversation data based on the intention database, and identify a target string by excluding the string portion containing the exclusion phrase from the extracted string portions based on the exclusion database. A specifying unit, An evaluation unit that evaluates the conversation data based on the target string, A conversation evaluation device comprising the same. (Appendix 2) The conversation evaluation device according to Appendix 1, wherein the exclusion phrase is a phrase including the end of a declarative sentence consisting of a string in which other characters are concatenated to the end of the intention phrase. Conversation evaluation device. (Appendix 3) The conversation evaluation device according to Appendix 2, wherein the exclusion phrase is a phrase including the end of the intention phrase. Conversation evaluation device. (Appendix 4) The conversation evaluation device according to Appendix 1, wherein the intention phrase is a phrase of any intention expression of confirmation, instruction, or request by the conversation participant to the conversation partner. Conversation evaluation device. (Appendix 5) The conversation evaluation device according to Appendix 1, wherein the evaluation unit evaluates the conversation data based on the appearance status of the target string in the conversation data. Conversation evaluation device. (Appendix 6) The conversation evaluation device according to Appendix 5, comprising a noun database in which preset nouns are registered, wherein the specifying unit identifies the nouns registered in the noun database from the obtained conversation data, and the evaluation unit evaluates the conversation data based on the appearance status of the nouns and the appearance status of the target string in the conversation data. Conversation evaluation device. (Appendix 7) The conversation evaluation device according to Appendix 6, The evaluation unit evaluates the conversation data based on the number of nouns that appear in the conversation data and the number of target character strings. Conversation evaluation device. (Appendix 8) A conversation evaluation device according to Appendix 6, The evaluation unit evaluates the conversation data based on the content of the nouns that appear in the conversation data and the content of the target character string. Conversation evaluation device. (Appendix 9) An intention database in which preset intention phrases, which are phrases for expressing intentions by a conversant, are registered, and An exclusion database in which preset exclusion phrases, which are phrases consisting of at least a part of declarative sentences including at least a part of the intention phrases, are registered, An information processing apparatus including extracts a character string portion including the intention phrase from the acquired conversation data based on the intention database, and identifies a target character string obtained by excluding the character string portion including the exclusion phrase from among the extracted character string portions based on the exclusion database, evaluates the conversation data based on the target character string. Conversation evaluation method. (Appendix 10) An intention database in which preset intention phrases, which are phrases for expressing intentions by a conversant, are registered, and An exclusion database in which preset exclusion phrases, which are phrases consisting of at least a part of declarative sentences including at least a part of the intention phrases, are registered, in an information processing apparatus including extracts a character string portion including the intention phrase from the acquired conversation data based on the intention database, and identifies a target character string obtained by excluding the character string portion including the exclusion phrase from among the extracted character string portions based on the exclusion database, evaluates the conversation data based on the target character string. A program for causing the execution of the process.

Explanation of Symbols

[0034] 1 User terminal 11 Acquisition unit 12 Analysis unit 13 Evaluation unit 16 Phrase storage unit 17 Call data storage unit

Claims

1. An intention database in which preset intention phrases, which are phrases of intention expressions by a speaker, are registered, An exclusion database in which preset exclusion phrases, which are phrases consisting of at least a part of a declarative sentence including at least a part of the intention phrase, are registered, A specifying unit that extracts a character string portion including the intention phrase based on the intention database from the acquired conversation data, and specifies a target character string obtained by excluding the character string portion including the exclusion phrase based on the exclusion database from among the extracted character string portions, An evaluation unit that evaluates the conversation data based on the target character string, A conversation evaluation device comprising the above.

2. The conversation evaluation device according to Claim 1, wherein the exclusion phrase is a phrase including the end of a declarative sentence consisting of a character string in which other characters are concatenated to the end of the intention phrase, A conversation evaluation device.

3. The conversation evaluation device according to Claim 2, wherein the exclusion phrase is a phrase including the end of the intention phrase, A conversation evaluation device.

4. The conversation evaluation device according to Claim 1, wherein the intention phrase is a phrase of an intention expression of any one of confirmation, instruction, or request by the speaker to the conversation partner, A conversation evaluation device.

5. The conversation evaluation device according to Claim 1, wherein the evaluation unit evaluates the conversation data based on the appearance status of the target character string in the conversation data, A conversation evaluation device.

6. The conversation evaluation device according to Claim 5, comprising a noun database in which preset nouns are registered, wherein the specifying unit specifies the nouns registered in the noun database from the acquired conversation data, and the evaluation unit evaluates the conversation data based on the appearance status of the nouns and the appearance status of the target character string in the conversation data, A conversation evaluation device.

7. The conversation evaluation device according to Claim 6, wherein the evaluation unit evaluates the conversation data based on the number of the nouns that appear in the conversation data and the number of the target character strings, A conversation evaluation device.

8. The conversation evaluation device according to Claim 6, wherein the evaluation unit evaluates the conversation data based on the content of the nouns that appear in the conversation data and the content of the target character string, A conversation evaluation device.

9. An intention database in which preset intention phrases, which are phrases of intention expressions by a speaker, are registered, An exclusion database in which a preset exclusion phrase, which is a phrase consisting of at least a part of a declarative sentence including at least a part of the intention phrase, is registered, An information processing apparatus including: extracts a character string portion including the intention phrase from the acquired conversation data based on the intention database, and identifies a target character string obtained by excluding the character string portion including the exclusion phrase from among the extracted character string portions based on the exclusion database; evaluates the conversation data based on the target character string; A conversation evaluation method. **Claim 10** An intention database in which a preset intention phrase, which is a phrase for expressing an intention by a conversant, is registered, An exclusion database in which a preset exclusion phrase, which is a phrase consisting of at least a part of a declarative sentence including at least a part of the intention phrase, is registered, An information processing apparatus including: extracts a character string portion including the intention phrase from the acquired conversation data based on the intention database, and identifies a target character string obtained by excluding the character string portion including the exclusion phrase from among the extracted character string portions based on the exclusion database; evaluates the conversation data based on the target character string; A program for causing the above processing to be executed.

Images