User assistance device and user assistance method

Abstract

This user assistance device is provided with: a query acquisition unit that acquires information expressed in a natural language that expresses query content; a feature vector calculation unit that calculates a feature vector for the query content; an answer source storage unit that stores at least one answer source for the query; an answer source retrieval unit that retrieves one or more answer sources from the answer source storage unit on the basis of the feature vector; a transmission unit that transmits an answer source of the query content and the search result to the dialogue-type response device; a reception unit that receives an answer from the dialogue-type response device; a summary unit that creates a summary of an answer source on the basis of the answer source of the search result; and a presentation unit that outputs the received answer and abstract text.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to a user assistance device and a user assistance method. BACKGROUND

[0002] In a manufacturing site such as a factory where industrial machines such as machines and robots are installed, as one of the methods for dealing with inquiries related to industrial machines, there is an inquiry response using a chat robot or the like (Patent Documents 1, 2, etc.). As one of the methods for realizing a chat robot that can deal with various inquiries from users, there is a method of using an LLM (Large Language Model). A chat robot using an LLM can appropriately answer inquiries using natural language.

[0003] PRIOR ART DOCUMENTS

[0004] PATENT DOCUMENTS

[0005] Patent Document 1: Japanese Patent Application Publication No. 2022-112541

[0006] Patent Document 2: Japanese Patent Application Publication No. 2019-160286 SUMMARY

[0007] PROBLEMS TO BE SOLVED BY THE INVENTION

[0008] The answer output by the LLM is not necessarily always correct. Therefore, in order to verify whether the answer is a correct answer, the user who obtained the answer exists a technology of inquiring the LLM in advance by limiting the source for generating the answer, and generating an embedding of the answer by the LLM on the basis of the source being made clear. However, in order to confirm the authenticity of the answer, the user needs to read the answer source carefully. This will be a burden for the user. Therefore, a technology in which the user can easily confirm whether the answer of the chat robot is accurate is desired.

[0009] MEANS FOR SOLVING THE PROBLEMS

[0010] The user assistance device of the present disclosure summarizes the answer source using a method capable of extracting accurate information from the answer of the LLM, and prompts the user with the answer of the LLM, thereby solving the above-mentioned problems.

[0011] Also, one embodiment of the present disclosure is a user assistance device including: a query acquisition unit that acquires information expressed in a natural language expressing a query content; a feature vector calculation unit that calculates a feature vector for the query content; a response source storage unit that stores at least one response source for a query; a response source search unit that searches for one or more response sources from the response source storage unit based on the feature vector; a transmission unit that transmits the query content and the searched response source to a dialog-type response device; a reception unit that receives a response from the dialog-type response device; an abstract creation unit that creates an abstract text of the searched response source based on the response source; and a presentation unit that outputs the received response and the abstract text. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 is a schematic hardware configuration diagram of a user assistance device of the first embodiment of the present disclosure.

[0013] Figure 2 is a block diagram showing the schematic functions of the user assistance device of the first embodiment.

[0014] Figure 3 is a diagram showing an example of a plurality of response sources stored in the response source storage unit.

[0015] Figure 4 is a screen diagram showing an example in which the presentation unit displays a response and an abstract text from the dialog-type response device.

[0016] Figure 5 is a block diagram showing the schematic functions of the abstract creation unit of the modified example. DETAILED DESCRIPTION

[0017] Hereinafter, with reference to the accompanying drawings, the embodiments of the present disclosure will be described. Figure 1 The embodiments of the present disclosure will be described below.

[0018] [First Embodiment]

[0019] Figure 1 is a schematic hardware configuration diagram showing the main part of the user assistance device of the first embodiment of the present disclosure. The user assistance device 1 of the present embodiment can be installed on, for example, a control device that controls an industrial machine according to a control program. In addition, it can be installed on a personal computer provided separately from the control device, a personal computer connected to the control device via a wired / wireless network, a cell computer, a fog computer 6, a cloud server 7, or the like. In the present embodiment, an example in which the user assistance device 1 is installed on a personal computer connected to a control device that controls an industrial machine via a network is shown.

[0020] The CPU 11 of the user assistance device 1 of the present embodiment is a processor that controls the entire user assistance device 1. The CPU 11 reads out a system program stored in the ROM 12 via a bus 22 and controls the entire user assistance device 1 in accordance with the system program. Temporary calculation data, display data, and various data acquired from the outside, and the like are temporarily stored in the RAM 13.

[0021] The nonvolatile memory 14 is constituted by, for example, a memory backed up by a battery not shown, an SSD (Solid State Drive), or the like, and retains a storage state even when the power of the user assistance device 1 is turned off. In the nonvolatile memory 14, programs, data read in from the external device 72 via the interface 15, programs, data input via the input device 71, programs, data acquired from the industrial machine 4, other devices via the network 5, and the like are stored. The programs, data stored in the nonvolatile memory 14 can also be expanded in the RAM 13 at the time of execution / use. In addition, various system programs such as a known analysis program are written in the ROM 12 in advance.

[0022] The interface 15 is an interface for connecting the CPU 11 of the user assistance device 1 and the external device 72 such as a USB device. From the external device 72 side, for example, a system program, setting data, and the like are read in. In addition, programs, setting data, and the like made and edited in the user assistance device 1 can be stored in an external storage unit via the external device 72.

[0023] The interface 20 is an interface for connecting the CPU 11 of the user assistance device 1 and the network 5 which is wired or wireless. The network 5 can communicate using, for example, serial communication such as RS-485, Ethernet (registered trademark) communication, optical communication, wireless LAN, Wi-Fi (registered trademark), Bluetooth (registered trademark), or the like. The network 5 is connected to the industrial machine 4 which is at least one control target, at least one other industrial machine 4, the dialogue-type response device 2 which responds to a query from a user, the fog computer 6, the cloud server 7, and the like, and data is exchanged between them and the user assistance device 1.

[0024] Each data read into the memory, data obtained as a result of executing a program, and the like are output via the interface 17 and displayed on the display device 70. In addition, the input device 71 constituted by a keyboard, a pointing device, and the like transmits instructions, data, and the like according to an operator's operation to the CPU 11 via the interface 18.

[0025] The dialogue-type response device 2 is a device that returns a response article in response to a prescribed article. The dialogue-type response device 2 has a large language model (LLM) that has learned a response article in response to a prescribed article. This model can use, for example, a known model such as a Transformer model. The large language model learns the probability of becoming a response to a prescribed article when a prescribed article is input for each text. In addition, the probability of continuing to other each text is further learned for the text. By repeatedly performing such learning, the large language model outputs a text string with a high probability as a response to the article when a prescribed article is input. In learning, the dialogue-type response device 2 can be used for the purpose of dialogue, question and answer, summary of an article, editing of an article, translation of an article, conversion of an article, change of an article, optimization of an article, explanation and detection of an article, recognition, prediction, determination, code generation, image generation, comprehensive determination, and the like, depending on how the document that becomes a response to a prescribed article is learned.

[0026] The user assistance device 1 of the present disclosure is premised on being connected to the dialogue-type response device 2 via the network 5. This dialogue-type response device 2 can also learn a response to a question based on machine information of the industrial machine 4. In learning of the dialogue-type response device 2, for example, a question actually performed to a consultation desk or the like and an answer thereto in the past are used. By reducing information for learning, the dialogue-type response device 2 that returns a response with a certain degree of accuracy in response to a question can be prepared.

[0027] Figure 2 is a diagram that shows the functions that the user assistance device 1 of the first embodiment of the present disclosure has as a rough block diagram. The user assistance device 1 of the present embodiment has each function shown in the diagram by the CPU 11 executing a system program and controlling the actions of each part of the user assistance device 1. Figure 1 The user assistance device 1 shown in the diagram has the CPU 11 executing a system program and controlling the actions of each part of the user assistance device 1 to realize.

[0028] The user assistance device 1 of the present embodiment has a question acquisition section 100, a feature vector calculation section 110, an answer source search section 120, a transmission section 130, a reception section 140, a summary section 150, and a prompting section 160. In addition, in the RAM 13 or the nonvolatile memory 14 of the user assistance device 1, an area that stores at least one answer source that includes information related to an answer to a question, that is, an answer source storage section 200, is prepared.

[0029] The inquiry acquisition unit 100 acquires inquiry information including inquiry content described in natural language. The inquiry acquisition unit 100, for example, can also cause the display device 70 to display a user interface for a user of the industrial machine 4 to input inquiry content, acquire inquiry content input by the user from the input device 71. In addition, the inquiry content can also be acquired from a voice input by the user from a voice input device such as a microphone not illustrated. Furthermore, the inquiry content from the user can also be input using a display device and an input device not illustrated that the industrial machine 4 has. The inquiry acquisition unit 100 outputs the acquired inquiry information to the feature vector calculation unit 110 and the transmission unit 130.

[0030] The feature vector calculation unit 110 calculates a feature vector for inquiry content including inquiry information acquired by the inquiry acquisition unit 100. The feature vector represents the characteristics of a prescribed article in vector form. For example, the calculation of the feature vector from the inquiry information can be delegated to the dialog-type response device 2. In addition, as the function of the feature vector calculation unit 110, a known large-scale language model such as a Transformer model can be caused to act, and the feature vector can be calculated using this model. In this case, the value of the intermediate layer when the article of the inquiry content is input to the model (the output value of the encoder portion) can be vectorized as the feature amount. Furthermore, a known technique such as a BOW (Bag Of Word) that represents an article in vector form using the frequency of occurrence of a word appearing in the article as an element, a TF-IDF (Term Frequency, Inverse Document Frequency) that vectorizes the frequency of occurrence of a word by assigning a weight (importance) to the frequency of occurrence of the word, or the like can be used to calculate the feature vector of the article. The feature vector calculation unit 110 outputs the calculated feature vector to the answer source search unit 120.

[0031] The answer source search unit 120 searches for one or more answer sources from the answer source storage unit 200 on the basis of the feature vector calculated by the feature vector calculation unit 110. The answer source search unit 120 calculates the degree of similarity between the feature vector for the inquiry content and the feature vector related to each answer source stored in the answer source storage unit 200, and the answer source related to the feature vector having a large degree of similarity is output as a search result. The answer source search unit 120 outputs the answer source as a search result to the transmission unit 130 and the summary unit 150.

[0032] Figure 3is a diagram showing an example of the answer source stored in the answer source storage section 200. At least one answer source is stored in the answer source storage section 200 in advance, which is obtained by dividing a technical document such as a manual of the industrial machine 4 in units of a certain number of characters (for example, 4000 characters). Each answer source is associated with a feature vector calculated by the same method as that for calculating the feature vector of the inquiry content. The answer source retrieval section 120 calculates the similarity between the feature vector x1, x2,... associated with each answer source 1, 2,... and the feature vector y = (b1, b2, b3,..., bn) of the inquiry content. For example, the similarity can be calculated using the distance between the vectors, or a formula in which the smaller the angle between the vectors, the larger the value. n )between the feature vector y = (b1, b2, b3,..., bn) of the inquiry content. For example, the similarity can be calculated using the distance between the vectors, or a formula in which the smaller the angle between the vectors, the larger the value.

[0033] The transmission section 130 transmits the inquiry information acquired by the inquiry acquisition section 100 and the answer source retrieved by the answer source retrieval section 120 to the dialog-type response device 2. The dialog-type response device 2 that receives the inquiry information and the answer source from the transmission section 130 generates an answer to the inquiry information. When generating the answer, the dialog-type response device 2 creates an answer based on the transmitted answer source. At this time, the dialog-type response device 2 acquires domain knowledge related to the inquiry information from the answer source, and generates an answer to the inquiry information after adding it to the inquiry information. Then, the generated response is transmitted to the user assistance device 1.

[0034] The reception section 140 receives the answer transmitted from the dialog-type response device 2. Then, the received answer is output to the summary section 150 and the prompting section 160.

[0035] The summary section 150 creates a summary text of the answer source based on the answer source of the retrieval result of the answer source retrieval section 120. The summary section 150 can create a summary text of the answer source by a publicly known method such as Graph Base Methods, Feature Base Methods, Topic Base Methods, and the like. The summary section 150 outputs the created summary text to the prompting section 160.

[0036] The prompting section 160 prompts the user with the answer received by the reception section 140 from the dialog-type response device 2 and the summary text created by the summary section 150. The prompting section 160 can also prompt the user with the answer and the summary text by, for example, arranging and displaying the answer and the summary text on the display device 70. At this time, the prompting section 160 prompts in a form in which the answer and the summary text can be distinguished. For example, the answer and the summary text can be displayed in different display fields, or the color, font, style, size, and the like of the text can be changed.

[0037] Figure 4is a screen diagram showing an example of the answer and the abstract text from the dialog-type response device 2 that the presentation section 160 presents. In Figure 4 In the example, the user asks a method of rotating the coordinate. In contrast, the abstract texts of the answers from the dialog-type response device 2 and the manual (answer source) related to the rotation of the coordinate are arranged and displayed. The user can view the abstract texts of the manual as needed while referring to the answers from the dialog-type response device 2. It is assumed that even if the answer from the dialog-type response device 2 is incorrect, by viewing the abstract texts of the manual, it is possible to easily confirm whether the answer is incorrect.

[0038] The user assistance device 1 of the present embodiment having the above-described structure displays the abstract texts of the answer sources similar to the inquiry in arrangement with the answer of the dialog-type response device 2 to the inquiry when the user makes the inquiry to the dialog-type response device 2. The user can view the abstract texts of the answer sources while referring to the answer of the dialog-type response device 2, and can easily confirm whether the answer of the dialog-type response device 2 is accurate. Therefore, it is possible to reduce the burden of the user at the time of the answer confirmation. Generally, the dialog-type response device 2 using the LLM makes an answer in natural articles, but sometimes there is a problem in the accuracy. On the other hand, the abstract texts of the answer sources such as the manual become slightly unnatural articles, but are mostly accurate within the range of the description. The user assistance device 1 of the present embodiment provides a user assistance function that plays these respective advantages.

[0039] As a modification example of the user assistance device 1 of the present embodiment, the abstract section 150 can further select the article for the abstract text from the answer sources of the search results. Figure 5 is a diagram showing the functions of the abstract section 150 of the user assistance device 1 of the present modification example as a schematic block diagram. The abstract section 150 of the present modification example has a division section 152, a score calculation section 154, and a selection section 156.

[0040] The division section 152 analyzes the article of the answer source searched by the answer source search section 120 and divides it into a prescribed article unit. The division section 152 can also divide the answer source into an article unit having a prescribed meaning such as a chapter, a section, and the like, for example. This division is used to divide the manual into an article of each technology unit, for example.

[0041] The score calculation section 154 calculates a score indicating the similarity to the response from the dialog-type response device 2 for each article unit divided by the division section 152. For example, the score indicating the similarity can be calculated from the ratio, the total number, and the like of the words included in the answer from the dialog-type response device 2 among the words included in the divided article unit. As long as the more words included in the answer from the dialog-type response device 2 are included in the words of the article unit, the higher the score.

[0042] The selection section 156 selects a prescribed article unit from among the plurality of word units segmented by the segmentation section 152 on the basis of the scores calculated by the score calculation section 154. The selection section 156 can select an article unit with the highest score. Also, the abstract section 150 creates an abstract text on the basis of the article unit selected by the selection section 156.

[0043] Further, the presentation section 160 of this modification example can also change the display of the abstract text on the basis of the scores calculated by the score calculation section 154 when the abstract text is displayed. For example, the highlighting of the characters or the like can be changed on the basis of the level of the scores, whereby the reliability of the abstract text can be expressed.

[0044] The user assistance device 1 of this modification example selects an article unit similar to the answer from the dialog-type response device 2 from among the answer sources and creates an abstract text on the basis of the article unit. Therefore, the user can easily confirm the accuracy of the answer while viewing the abstract text that has been more accurately abstracted.

[0045] The above describes the embodiments of the present disclosure in detail, but the present disclosure is not limited to the above-described respective embodiments. The embodiments can be variously added, replaced, changed, partially deleted, or the like within a range not departing from the gist of the invention, or within a range not departing from the idea and gist of the present disclosure derived from the content described in the claims and equivalents thereof. For example, in the above-described embodiments, the order of the respective actions, the order of the respective processes are indicated as an example, and are not limited thereto. The same applies to the case where numerical values or mathematical expressions are used in the description of the above-described embodiments.

[0046] Hereinafter, the postscript of the embodiments of the present disclosure is described.

[0047] (Postscript 1)

[0048] The user assistance device (1) of one embodiment of the present disclosure includes an inquiry acquisition section (100) that acquires information expressed in a natural language expressing an inquiry content, a feature vector calculation section (110) that calculates a feature vector for the inquiry content, an answer source storage section (200) that stores at least one answer source for an inquiry, an answer source search section (120) that searches for one or more answer sources from the answer source storage section (200) on the basis of the feature vector, a transmission section (130) that transmits the inquiry content and the searched answer source to a dialog-type response device (2), a reception section (140) that receives an answer from the dialog-type response device (2), an abstract section (150) that creates an abstract text of the searched answer source on the basis of the answer source, and a presentation section (160) that outputs the received answer and the abstract text.

[0049] (Postscript 2)

[0050] The summary section (150) of the other user assistance device (1) of the present disclosure has a division section (152) that divides the answer sources of the search results, a score calculation section (154) that assigns scores to the results of the division using the answer pairs received from the dialog-type response device (2), and a selection section (156) that selects based on the scores.

[0051] (Addendum 3)

[0052] The answer source storage section (200) of the other user assistance device (1) of the present disclosure stores text of a certain length and a feature vector for each text together, and the answer source search section (120) searches based on the distance between the feature vector of the question content and the feature vectors of the answer sources.

[0053] (Addendum 4)

[0054] The prompt section (160) of the other user assistance device (1) of the present disclosure outputs in a form that can distinguish between the received content and the summary content.

[0055] (Addendum 5)

[0056] The prompt section (160) of the other user assistance device (1) of the present disclosure changes the output content based on the scores calculated by the summary section (150).

[0057] (Addendum 6)

[0058] Regarding the other user assistance method of the present disclosure, the following steps are performed by a computer: a step of obtaining information expressed in a natural language that expresses a question content; a step of calculating a feature vector for the question content; a step of searching for at least one answer source from an answer source storage section (200) that stores at least one answer source for a question based on the feature vector; a step of sending the question content and the answer sources of the search results to a dialog-type response device (2); a step of receiving an answer from the dialog-type response device (2); a step of creating a summary text of the answer sources based on the answer sources of the search results; and a step of outputting the received answer and the summary text.

[0059] Explanation of Reference Signs

[0060] 1 User Assistance Device

[0061] 2 Dialog-Type Response Device

[0062] 4 Industrial Machine

[0063] 5 Network

[0064] 6 Fog Computer

[0065] 7 cloud server

[0066] 11 CPU

[0067] 12 ROM

[0068] 13 RAM

[0069] 14 nonvolatile memory

[0070] 15, 17, 18, 20 interface

[0071] 22 bus

[0072] 70 display device

[0073] 71 input device

[0074] 72 external device

[0075] 100 inquiry acquisition section

[0076] 110 feature vector calculation section

[0077] 120 answer source search section

[0078] 130 transmission section

[0079] 140 reception section

[0080] 150 summarization section

[0081] 152 division section

[0082] 154 score calculation section

[0083] 156 selection section

[0084] 160 prompting section

[0085] 200 answer source storage section

Claims

1. A user assistance device, characterized in that, have: The inquiry and retrieval department retrieves information expressed in natural language that conveys the content of the inquiry. A feature vector calculation unit calculates a feature vector for the query content; The response source storage unit stores at least one response source for each query. The answer source retrieval unit retrieves one or more answer sources from the answer source storage unit based on the feature vector. The sending unit sends the query content and the source of the response to the search results to the conversational response device; The receiving unit receives responses from the dialog-type response device; The abstracting department, based on the source of the response in the search results, creates a summary of the response source; The prompting unit outputs the received answer and the summary text.

2. The user assistance device according to claim 1, characterized in that, The abstract section has: A segmentation unit that segments the source of the search results; The scoring unit assigns a score to the segmentation result using the response received from the dialogic response device; The selection section makes selections based on the scores.

3. The user assistance device according to claim 1, characterized in that, The answer source storage unit stores a certain length of text along with the feature vectors for each text. The answer source retrieval unit performs the retrieval based on the distance between the feature vector of the query content and the feature vector of the answer source.

4. The user assistance device according to claim 1, characterized in that, The prompt section outputs information in a format that distinguishes between received content and summary content.

5. The user assistance device according to claim 2, characterized in that, The prompting section changes the output content based on the score calculated by the summary section.

6. A user assistance method, characterized in that, Executed by computer: The steps to obtain information expressed in natural language that represents the content of the inquiry; The step of calculating the feature vector for the query content; The step of retrieving at least one of the answer sources from an answer source storage unit that stores at least one answer source for a query, based on the feature vector; The step of sending the query content and the source of the search results to the conversational response device; The step of receiving a response from the dialogic response device; The steps for creating a summary of the response source based on the search results; The steps include outputting the received response and the summary text.

Images