User assistance device and user assistance method

Abstract

A user assistance device according to the present disclosure comprises: an inquiry acquisition unit that acquires information which is expressed in a natural language expressing the details of an inquiry; a feature vector calculation unit that calculates a feature vector for the details of the inquiry; an answer source storage unit that stores at least one answer source corresponding to the inquiry; 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, to an interactive response device, the details of the inquiry and the answer source of the search result; a reception unit that receives the answer from the interactive response device; a summarization unit that creates a summary sentence of the answer source on the basis of the answer source of the search result; and a presentation unit that outputs the received answer and the summary sentence.

Description

User assistance device and user assistance method

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

[0002] In manufacturing sites such as factories where industrial machines such as machine tools and robots are installed, one method for responding to inquiries about the industrial machines is to use a chatbot (see, for example, Patent Literature 1 and 2). One method for realizing a chatbot that can respond to various inquiries from users is to use an LLM (Large Language Model). A chatbot using an LLM can appropriately respond to inquiries in natural language.

[0003] JP 2022-112541 A JP 2019-160286 A

[0004] The answers output by an LLM are not always correct. Therefore, in order for a user who receives an answer to verify whether the answer is correct, a technique called embedding is available, in which the source for generating the answer is limited in advance, an inquiry is made to the LLM, and the LLM generates an answer after the source is made clear. However, to verify the authenticity of the answer, the user must carefully read the answer source. This is a burden for the user. Therefore, there is a need for a technique that allows users to easily verify whether the chatbot's answer is accurate.

[0005] The user assistance device according to the present disclosure solves the above problem by summarizing the answer source based on the LLM answer using a method capable of extracting accurate information and presenting it to the user together with the LLM answer.

[0006] One aspect of the present disclosure is a user assistance device comprising: a query acquisition unit that acquires information expressed in a natural language that expresses the 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 search unit that searches for one or more answer sources from the answer source storage unit based on the feature vector; a transmission unit that transmits the query content and the answer sources of the search results to an interactive response device; a reception unit that receives answers from the interactive response device; a summarization unit that creates a summary of the answer sources based on the answer sources of the search results; and a presentation unit that outputs the received answers and the summary.

[0007] [Correction based on Rule 91 01.10.2025] Fig. 1 is a schematic hardware configuration diagram of a user assistance device according to a first embodiment of the present disclosure. Fig. 2 is a block diagram showing the general functions of the user assistance device according to the first embodiment. Fig. 3 is a schematic diagram showing an example of a plurality of answer sources stored in an answer source storage unit. Fig. 4 is a screen diagram showing an example in which a presentation unit displays an answer from an interactive response device and a summary sentence. Fig. 5 is a block diagram showing the general functions of a summary unit according to a modified example.

[0008] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. [First Embodiment] Fig. 1 is a schematic hardware configuration diagram showing the main parts of a user assistance device according to a first embodiment of the present disclosure. The user assistance device 1 according to this embodiment can be implemented, for example, on a control device that controls industrial machinery based on a control program. It can also be implemented on a computer such as a personal computer attached to the control device, or a personal computer, cell computer, fog computer 6, or cloud server 7 connected to the control device via a wired or wireless network. In this embodiment, an example is shown in which the user assistance device 1 is implemented on a personal computer connected to a control device that controls industrial machinery via a network.

[0009] The CPU 11 provided in the user assistance device 1 according to this embodiment is a processor that controls the entire user assistance device 1. The CPU 11 reads a system program stored in the ROM 12 via the bus 22 and controls the entire user assistance device 1 in accordance with the system program. The RAM 13 temporarily stores temporary calculation data, display data, and various data acquired from outside.

[0010] The non-volatile memory 14 is composed of, for example, a battery-backed memory (not shown) or an SSD (Solid State Drive), and retains its stored state even when the user assistance device 1 is powered off. The non-volatile memory 14 stores programs and data read from an external device 72 via the interface 15, programs and data input via the input device 71, and programs and data acquired from the industrial machine 4 or other devices via the network 5. The programs and data stored in the non-volatile memory 14 may be expanded into the RAM 13 when executed / used. In addition, various system programs, such as known analysis programs, are pre-written in the ROM 12.

[0011] The interface 15 is an interface for connecting the CPU 11 of the user assistance device 1 to an external device 72 such as a USB device. For example, system programs, setting data, etc. are read from the external device 72. In addition, programs and setting data created and edited within the user assistance device 1 can be stored in external storage means via the external device 72.

[0012] The interface 20 is an interface for connecting the CPU 11 of the user assistance device 1 to a wired or wireless network 5. The network 5 may communicate using technologies such as serial communication such as RS-485, Ethernet (registered trademark), optical communication, wireless LAN, Wi-Fi (registered trademark), Bluetooth (registered trademark), etc. At least one industrial machine 4 to be controlled, at least one other industrial machine 4, an interactive response device 2 that responds to inquiries from users, a fog computer 6, a cloud server 7, etc. are connected to the network 5, and data is exchanged between the network 5 and the user assistance device 1.

[0013] The display device 70 displays various data read into the memory, data obtained as a result of executing programs, etc., output via the interface 17. An input device 71, which is comprised of a keyboard, pointing device, etc., passes instructions, data, etc. based on operations by an operator to the CPU 11 via the interface 18.

[0014] The interactive response device 2 is configured as a device that returns a response sentence in response to a predetermined sentence. The interactive response device 2 includes a large-scale language model (LLM) that has learned a response sentence for the predetermined sentence. This model can be a known model such as a Transformer model. The large-scale language model learns, for each text, the probability that the given sentence will be a response to the given sentence when the given sentence is input. It also learns the probability that each of the given texts will be followed by other texts. By repeating this learning process, the large-scale language model, when input, comes to output a text string with a high probability as a response to the given sentence. Depending on how the learning process learns documents that will be responses to the given sentence, the interactive response device 2 can be used for purposes such as dialogue, question and answer session, text summarization, text editing, text translation, text conversion, text modification, text optimization, text interpretation / detection, recognition, prediction, judgment, code generation, image generation, and comprehensive judgment.

[0015] It is assumed that the user assistance device 1 according to the present disclosure is connected to an interactive response device 2 via a network 5. This interactive response device 2 may be one that has learned responses to inquiries based on machine information of an industrial machine 4. In learning the interactive response device 2, for example, inquiries that have actually been made to a help desk in the past and the responses to those inquiries are used. By narrowing down the information used for learning, it is possible to prepare an interactive response device 2 that returns responses to inquiries with a certain degree of accuracy.

[0016] 2 is a schematic block diagram showing functions of the user assistance device 1 according to the first embodiment of the present disclosure. Each function of the user assistance device 1 according to this embodiment is realized by the CPU 11 of the user assistance device 1 shown in FIG. 1 executing a system program and controlling the operation of each part of the user assistance device 1.

[0017] The user assistance device 1 of this embodiment includes a query acquisition unit 100, a feature vector calculation unit 110, an answer source search unit 120, a transmission unit 130, a reception unit 140, a summary unit 150, and a presentation unit 160. The RAM 13 to the non-volatile memory 14 of the user assistance device 1 also include an answer source storage unit 200, which is an area that stores in advance at least one answer source that includes information related to an answer to a query.

[0018] The query acquisition unit 100 acquires query information including query content written in natural language. The query acquisition unit 100 may, for example, display a user interface on the display device 70 for the user of the industrial machine 4 to input query content, and acquire the query content input by the user from the input device 71. Alternatively, the query content may be acquired from a voice uttered by the user input from a voice input device such as a microphone (not shown). Furthermore, the query content from the user may be input using a display device and an input device (not shown) provided in the industrial machine 4. The query acquisition unit 100 outputs the acquired query information to the feature vector calculation unit 110 and the transmission unit 130.

[0019] The feature vector calculation unit 110 calculates a feature vector for the query content, including the query information acquired by the query acquisition unit 100. The feature vector represents the features of a given sentence in vector format. For example, the interactive response system 2 may be requested to calculate a feature vector based on the query information. Alternatively, a known large-scale language model, such as a Transformer model, may be operated as a function of the feature vector calculation unit 110, and the feature vector may be calculated using the model. In this case, the intermediate layer values ​​(output values ​​of the encoder) obtained when the query content sentence is input to the model may be vectorized as features. Furthermore, the feature vector of a sentence may be calculated using known techniques, such as BOW (Bag of Words), which represents a vector with elements representing the frequency of occurrence of words appearing in the sentence, or TF-IDF (Term Frequency, Inverse Document Frequency), which vectorizes the frequency of occurrence of words by weighting (importance). The feature vector calculation unit 110 outputs the calculated feature vector to the answer source search unit 120 .

[0020] The answer source search unit 120 searches for one or more answer sources from the answer source storage unit 200 based on the feature vector calculated by the feature vector calculation unit 110. The answer source search unit 120 calculates the similarity between the feature vector for the query content and the feature vectors associated with each answer source stored in the answer source storage unit 200, and determines the answer source associated with the feature vector having the greatest similarity as the search result. The answer source search unit 120 outputs the searched answer sources to the transmission unit 130 and the summarization unit 150.

[0021] FIG. 3 is a schematic diagram showing an example of an answer source stored in the answer source storage unit 200. The answer source storage unit 200 stores at least one answer source, which is obtained by dividing technical documents such as manuals for industrial machines 4 into units of a certain number of characters (for example, 4000 characters). Each answer source is associated with a feature vector calculated in the same way as the feature vector calculation unit 110 calculates the feature vector. The answer source search unit 120 searches for the feature vectors x1, x2, ... associated with each answer source 1, 2, ..., and finds a feature vector y=(b1, b2, b3, ..., b n For example, the similarity can be calculated using a formula that increases the value as the distance between vectors or the angle between vectors decreases.

[0022] The transmitting unit 130 transmits the inquiry information acquired by the inquiry acquiring unit 100 and the answer source searched by the answer source searching unit 120 to the interactive response device 2. The interactive response device 2, which has received the inquiry information and the answer source from the transmitting unit 130, generates an answer to the inquiry information. When generating an answer, the interactive response device 2 creates an answer based on the sent answer source. At this time, the interactive response device 2 acquires domain knowledge that is likely to be related to the inquiry information from the answer source, adds it to the inquiry information, and generates an answer to the inquiry information. The generated answer is then transmitted to the user assistance device 1.

[0023] The receiving unit 140 receives the answer sent from the interactive response apparatus 2. Then, the receiving unit 140 outputs the received answer to the summarizing unit 150 and the presenting unit 160.

[0024] The summarization unit 150 creates a summary of the answer source based on the answer source search results by the answer source search unit 120. The summarization unit 150 may create a summary of the answer source using known methods such as Graph-Based Methods, Feature-Based Methods, or Topic-Based Methods. The summarization unit 150 outputs the created summary to the presentation unit 160.

[0025] The presentation unit 160 presents to the user the answer received by the receiving unit 140 from the interactive response device 2 and the summary created by the summarizing unit 150. The presentation unit 160 may present the answer and the summary to the user by, for example, displaying the answer and the summary side by side on the display device 70. At this time, the presentation unit 160 presents the answer and the summary in a manner that allows them to be distinguished. For example, they may be displayed in different display fields, or may be displayed with different text colors, fonts, formats, sizes, etc.

[0026] FIG. 4 is a screen diagram showing an example of an answer and a summary from the interactive response apparatus 2 presented by the presenting unit 160. In the example of FIG. 4, the user inquires about a method for rotating coordinates. In response to this, the answer from the interactive response apparatus 2 and a summary from a manual (answer source) related to the rotation of coordinates are displayed side by side. The user can refer to the answer from the interactive response apparatus 2 and can also view the summary from the manual as needed. Even if the answer from the interactive response apparatus 2 contains an error, the user can easily check whether the answer is correct by looking at the summary from the manual.

[0027] When a user makes a query to the interactive response device 2, the user assistance device 1 according to this embodiment, having the above configuration, displays a summary of an answer source similar to the query alongside the answer from the interactive response device 2. The user can view the summary of the answer source while referring to the answer from the interactive response device 2, and can easily confirm whether the answer from the interactive response device 2 is accurate. This reduces the burden on the user when checking the answer. Generally, an interactive response device 2 using an LLM responds in natural sentences, but there may be problems with their accuracy. On the other hand, summaries from answer sources such as manuals tend to be somewhat unnatural, but are often accurate within the scope of the description. The user assistance device 1 according to this embodiment provides a user assistance function that makes use of the advantages of each of these.

[0028] In one modification of the user assistance device 1 according to this embodiment, the summarizing unit 150 may further select sentences to be used in the summary from the answer sources of the search results. Figure 5 is a schematic block diagram showing the functions of the summarizing unit 150 of the user assistance device 1 according to this modification. The summarizing unit 150 according to this modification includes a dividing unit 152, a score calculating unit 154, and a selecting unit 156.

[0029] The dividing unit 152 analyzes the sentences of the answer sources searched by the answer source searching unit 120 and divides them into predetermined sentence units. The dividing unit 152 may divide the answer sources into predetermined meaningful sentence units, such as chapters or sections. This division is intended to divide, for example, a manual into sentences for each technology unit.

[0030] The score calculation unit 154 calculates a score indicating the similarity between each sentence unit divided by the dividing unit 152 and the response from the interactive response system 2. This score indicating similarity may be calculated based on, for example, the proportion or total number of words contained in the divided sentence units that are contained in the response from the interactive response system 2. The more words contained in the sentence unit that are contained in the response from the interactive response system 2, the higher the score should be.

[0031] The selection unit 156 selects a predetermined sentence unit from the plurality of sentence units divided by the division unit 152 based on the scores calculated by the score calculation unit 154. The selection unit 156 may select the sentence unit with the highest score. Then, the summarization unit 150 creates a summary based on the sentence units selected by the selection unit 156.

[0032] In this modified example, the presentation unit 160 may change the display of the summary depending on the score calculated by the score calculation unit 154. For example, the reliability of the summary may be expressed by changing the highlighting of the characters depending on the score.

[0033] The user assistance device 1 of this modification selects sentence units similar to the answer from the interactive response device 2 from the answer source and creates a summary based on the sentence units. This allows the user to easily check the accuracy of the answer while looking at the summary that has been more precisely summarized.

[0034] Although the embodiments of the present disclosure have been described in detail above, the present disclosure is not limited to the individual embodiments described above. Various additions, substitutions, modifications, partial deletions, etc. are possible in these embodiments without departing from the gist of the invention or the idea and intent of the present disclosure derived from the content described in the claims and their equivalents. For example, in the above-described embodiments, the order of each operation and the order of each process are shown as examples and are not limited to these. The same applies when numerical values ​​or mathematical expressions are used in the description of the above-described embodiments.

[0035] The following are supplementary notes related to embodiments of the present disclosure. (Supplementary Note 1) A user assistance device (1) according to one aspect of the present disclosure includes a query acquisition unit (100) that acquires information expressed in a natural language that expresses a query content, a feature vector calculation unit (110) that calculates a feature vector for the query content, an answer source storage unit (200) that stores at least one answer source for the query, an answer source search unit (120) that searches for one or more answer sources from the answer source storage unit (200) based on the feature vector, a transmission unit (130) that transmits the query content and the answer sources of the search results to an interactive response device (2), a reception unit (140) that receives an answer from the interactive response device (2), a summarization unit (150) that creates a summary of the answer source based on the answer source of the search results, and a presentation unit (160) that outputs the received answer and the summary.

[0036] (Supplementary Note 2) The summarization unit (150) included in the user assistance device (1) according to another aspect of the present disclosure includes a division unit (152) that divides the answer sources of the search results, a score calculation unit (154) that assigns a score to the divided results using the answers received from the interactive response device (2), and a selection unit (156) that makes a selection based on the score. (Supplementary Note 3) The answer source storage unit (200) included in the user assistance device (1) according to another aspect of the present disclosure stores texts of a certain length together with feature vectors for each text, and the answer source search unit (120) performs a search based on the distance between the feature vector of the query content and the feature vector of the answer source.

[0037] (Supplementary Note 4) The presentation unit (160) included in the user assistance device (1) according to another aspect of the present disclosure outputs the received content and the summarized content in a manner that allows them to be distinguished. (Supplementary Note 5) The presentation unit (160) included in the user assistance device (1) according to another aspect of the present disclosure changes the output content depending on the score calculated by the summarization unit (150).

[0038] (Supplementary Note 6) A user assistance method according to another aspect of the present disclosure includes the steps of: acquiring information expressed in a natural language that expresses a query content; calculating a feature vector for the query content; searching for at least one answer source from an answer source storage unit (200) that stores at least one answer source for the query based on the feature vector; transmitting the query content and the answer source of the search result to an interactive response device (2); receiving an answer from the interactive response device (2); creating a summary of the answer source based on the answer source of the search result; and outputting the received answer and the summary.

[0039] REFERENCE SIGNS LIST 1 User assistance device 2 Interactive response device 4 Industrial machine 5 Network 6 Fog computer 7 Cloud server 11 CPU 12 ROM 13 RAM 14 Non-volatile memory 15, 17, 18, 20 Interface 22 Bus 70 Display device 71 Input device 72 External device 100 Query acquisition unit 110 Feature vector calculation unit 120 Answer source search unit 130 Transmission unit 140 Reception unit 150 Summarization unit 152 Division unit 154 Score calculation unit 156 Selection unit 160 Presentation unit 200 Answer source storage unit

Claims

1. A user assistance device comprising: a query acquisition unit that acquires information expressed in a natural language that expresses a 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 search unit that searches for one or more answer sources from the answer source storage unit based on the feature vector; a transmission unit that transmits the query content and the answer sources of the search results to an interactive response device; a reception unit that receives answers from the interactive response device; a summarization unit that creates a summary of the answer sources based on the answer sources of the search results; and a presentation unit that outputs the received answers and the summary.

2. The user assistance device of claim 1, wherein the summarizing unit comprises: a dividing unit that divides the answer sources of the search results; a score calculation unit that assigns a score to the divided results using the answers received from the interactive response device; and a selection unit that makes a selection based on the score.

3. The user assistance device of claim 1, wherein the answer source storage unit stores a certain length of text together with a feature vector for each text, and the answer source search unit performs a search 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, wherein the presentation unit outputs the received content and the summary content in a manner that allows them to be distinguished from each other.

5. The user assistance device according to claim 2, wherein the presentation unit changes output content according to the score calculated by the summarization unit.

6. A user assistance method in which a computer executes the steps of: acquiring information expressed in a natural language that expresses a query content; calculating a feature vector for the query content; searching for at least one answer source from an answer source storage unit that stores at least one answer source for the query based on the feature vector; transmitting the query content and the answer source of the search result to an interactive response device; receiving an answer from the interactive response device; creating a summary of the answer source based on the answer source of the search result; and outputting the received answer and the summary.

Images