USER SUPPORT DEVICE AND USER SUPPORT PROCEDURES

Abstract

A user support device according to the present disclosure comprises a query acquisition unit that obtains information expressed in natural language that reflects the details of a query; a feature vector calculation unit that calculates a feature vector for the details of the query; a response source storage unit that stores at least one response source corresponding to the 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 details of the query and the response source of the search result to an interactive response device; a receiving unit that receives the response from the interactive response device; and a summarization unit that generates a summary set of the response source based on the response source of the search result.and a presentation unit that displays the received answer and the summary sentence.

Description

TECHNICAL AREA

[0001] The present disclosure relates to a user support device and a user support method. GENERAL STATE OF THE ART

[0002] In production facilities such as factories equipped with industrial machinery like machine tools and robots, a method for responding to inquiries related to these machines is the use of a chatbox (see patent documents 1 and 2, etc.). One method for implementing a chatbox capable of responding to various user queries involves the use of a Large Language Model (LLM). The chatbox utilizing the LLM can respond appropriately to queries using natural language. LITERATURE LIST PATENT DOCUMENTS Patent document 1: JP 2022 - 112 541 A Patent document 2: JP 2019 - 160 286 A REVELATION OF THE INVENTION PROBLEM THAT THE INVENTION IS SUPPOSED TO SOLVE

[0003] Answers provided by the LLM (Language Learning Module) are not always correct. To allow users to verify the accuracy of a received response, a technology called embedding is used. This involves restricting the sources for generating an answer before querying the LLM, and then prompting the LLM to generate an answer only after the sources have been clearly identified. However, to verify the answer's accuracy, the user must carefully review the answer's source code. This is a burden for the user. Therefore, there is a need for a technology that allows users to easily verify the reliability of a chatbox response. MEANS TO SOLVENT THE PROBLEM

[0004] The user support device according to the disclosure solves the above problem by summarizing a response source using a method that is able to extract accurate information based on a response from an LLM and show the user the summary along with the response from the LLM.

[0005] One aspect of the disclosure is a user support device comprising a query acquisition unit configured to obtain information expressed in natural language that reflects the query content, a feature vector computation unit configured to compute a feature vector for the query content, a response source storage unit configured to store at least one response source for a query, a response source search unit configured to search for one or more response sources from the response source storage unit based on the feature vector, a transmission unit configured to transmit the query content and the response source resulting from the search to an interactive response device, and a receiving unit configured to receive a response from the interactive response device.a summary unit configured to generate a summary of the answer source based on the answer source obtained from the search, and a presentation unit configured to output the received answer and the summary. BRIEF DESCRIPTION OF THE DRAWINGS Fig. Figure 1 is a schematic hardware diagram of a user support device according to a first embodiment of the disclosure; Fig. 2 is a block diagram showing schematic functions of a user support device according to the first embodiment; Fig. Figure 3 is a schematic view showing an example of multiple response sources stored in a response source storage unit; Fig. Figure 4 is a screen view showing an example where a presentation unit displays a response from an interactive response device and a summary; and Fig. Figure 5 is a block diagram showing schematic functions of a summary unit according to a modified example. DETAILED DESCRIPTION

[0006] With reference to the drawings, embodiments of the disclosure will be described below. First embodiment

[0007] Fig. Figure 1 is a schematic hardware layout diagram showing the main components of a user support device according to a first embodiment of the disclosure. The user support device 1 according to this embodiment can, for example, be set up in a controller that controls an industrial machine based on a control program. Furthermore, the user support device 1 can be set up in a computer such as a personal computer installed together with the controller, a personal computer connected to the controller via a wired / wireless network, a cell computer, a fog computer 6, or a cloud server 7. This embodiment shows an example in which the user support device 1 is set up in a personal computer connected via the network to the controller that controls the industrial machine.

[0008] A CPU 11, with which the user support device 1 is equipped according to this embodiment, is a processor that controls the user support device 1 as a whole. The CPU 11 reads a system program stored in a ROM 12 via a bus 22 and controls the entire user support device 1 according to the system program. A RAM 13 temporarily stores computational data, display data, various data obtained from external sources, etc.

[0009] Non-volatile memory 14 includes, for example, memory supported by a battery (not shown), an SSD (solid-state drive), etc., and retains its memory state even when the power supply to the user support device 1 is switched off. The non-volatile memory 14 stores programs and data read from an external device 72 via an interface 15, programs and data entered through an input device 71, programs and data obtained from the industrial machine 4 or other devices via a network 5, etc. The programs and data stored in the non-volatile memory 14 can be loaded into RAM 13 during execution / use. In addition, various system programs, such as well-known analysis programs, are pre-written in ROM 12.

[0010] Interface 15 is used to connect the CPU 11 of the user support device 1 to the external device 72, such as a USB device. System programs, configuration data, etc., are read from the external device 72. Furthermore, programs, configuration data, etc., created or modified in the user support device 1 can be saved to an external storage device via the external device 72.

[0011] Interface 20 is an interface for connecting the CPU 11 of the user support device 1 to the wired or wireless network 5. The network 5 can communicate using a technology such as serial communication like RS-485, Ethernet (registered trademark), optical communication, wireless LAN, Wi-Fi (registered trademark), or Bluetooth (registered trademark). At least one industrial machine 4 to be controlled, at least one other industrial machine 4, an interactive response device 2 that responds to a request from a user, the fog computer 6, the cloud server 7, etc., are connected to the network 5 and exchange data with the user support device 1.

[0012] Various data written to memory, data obtained as a result of program execution, etc., are output via an interface 17 to a display device 70 and displayed there. Furthermore, the input device 71, which includes a keyboard, a pointing device, etc., delivers commands, data, etc., based on operator input, to the CPU 11 via an interface 18.

[0013] The Interactive Response Device 2 is configured as a device that provides a response set to a given sentence. The Interactive Response Device 2 incorporates a large language model (LLM) that has been trained on response sets to given sentences. This model may employ a known model such as a transformer model. The LLM is trained for each text using a probability that a response to a given sentence will be obtained when the given sentence is input. Furthermore, the LLM is trained using the possibility that each text is followed by another text. By repeating this training, the LLM, upon input of a given text, outputs a sequence of texts that has a high probability of being the response to the sentence.During training, the interactive response device 2 can be used for purposes such as dialogue, question answering, sentence summarization, sentence editing, sentence translation, sentence conversion, sentence modification, sentence optimization, sentence interpretation / detection, recognition, prediction, determination, code generation, image generation, and synthetic assessment, depending on how a document serving as a response to a given sentence is used for training.

[0014] It is assumed that, according to the disclosure, the user support device 1 is connected to the interactive response device 2 via network 5. The interactive response device 2 can be trained to respond to a query based on machine information from the industrial machine 4. For example, the training of the interactive response device 2 uses queries and responses that were actually made to a help desk in the past. By limiting the information used for training, an interactive response device 2 can be provided that returns a response to a query with a certain degree of accuracy.

[0015] Fig. Figure 2 shows functions of the user support device 1 according to the first embodiment of the disclosure as a schematic block diagram. Each function of the user support device 1 according to this embodiment is implemented by the CPU 11, with which the user support device 1, which is in Fig. 1 is shown, which executes a system program and controls the operation of each unit of the user support device 1.

[0016] The user support device 1 of this embodiment comprises a query acquisition unit 100, a feature vector calculation unit 110, an answer source search unit 120, a transmission unit 130, a receiving unit 140, a summarization unit 150, and a presentation unit 160. Furthermore, an answer source storage unit 200 is provided in the RAM 13 or the non-volatile memory 14 of the user support device 1. This answer source storage unit is an area in which at least one answer source containing information related to a query response has been pre-stored.

[0017] The query acquisition unit 100 obtains query information containing query content written in natural language. For example, the query acquisition unit 100 can display a user interface for entering query content to a user of industrial machine 4 on the display device 70 and obtain the query content entered by the user via the input device 71. The query content can also be obtained from a spoken utterance by the user, entered via a speech input device such as a microphone (not shown). Furthermore, the query content can be entered by the user using a display device and an input device (not shown) integrated into the industrial machine. The query acquisition unit 100 outputs the obtained query information to the feature vector calculation unit 110 and the transmission unit 130.

[0018] The feature vector calculation unit 110 calculates a feature vector for the query content, which contains the query information obtained by the query acquisition unit 110. The feature vector represents features of a given set in a vector format. For example, the interactive response device 2 can be required to calculate a feature vector based on the query information. Furthermore, as a function of the feature vector calculation unit 110, a known LLM, such as a transformer model, can be operated to calculate the feature vector using the model. In this case, an intermediate layer value (an output value of an encoder part) can be vectorized as a feature quantity when a query content set is input into the model.Furthermore, a feature vector of a sentence can be calculated using known techniques such as BOW (Bag of Words), which represents a vector containing as elements the frequencies of occurrence of words that appear in a sentence, or TF-IDF (Term Frequency, Inverse Document Frequency), which vectorizes the frequencies of occurrence of words by assigning them weights (a meaning). The feature vector calculation unit 110 outputs the calculated feature vector to the answer source search unit 120.

[0019] The answer source search unit 120 searches for one or more answer sources from the answer source storage unit 200 based on a feature vector calculated by the feature vector calculation unit 110. The answer source search unit 120 calculates the similarity between a feature vector representing the query content and a feature vector associated with each answer source stored in the answer source storage unit 200, and determines as the search result an answer source associated with a feature vector exhibiting a high degree of similarity. The answer source search unit 120 outputs the answer source representing the search result to the transmission unit 130 and the summarization unit 150.

[0020] Fig. Figure 3 is a schematic view illustrating an example of answer sources stored in the answer source storage unit 200. The answer source storage unit 200 contains at least one answer source in advance, generated by dividing a technical document, such as a manual for industrial machine 4, into units with a specific number of characters (e.g., 4000 characters). Each answer source is associated with a feature vector calculated in the same way as the feature vector calculation unit 110 calculates a feature vector. The answer source search unit 120 calculates the similarity between feature vectors x1, x2, ..., each associated with answer sources 1, 2, ..., and a feature vector y = (b1, b2, b3, ..., b n) for the query content. For example, similarity can be calculated using a formula where a value increases with decreasing distance or angle between vectors.

[0021] The transmission unit 130 transmits the query information, obtained by the query acquisition unit 100, and the response source, located by the response source search unit 120, to the interactive response device 2. Having received the query information and response source from the transmission unit 130, the interactive response device 2 generates a response to the query information. In generating a response, the interactive response device 2 creates the response based on the received response source. In this case, the interactive response device 2 obtains domain knowledge from the response source, which is likely related to the query information, adds this domain knowledge to the query information, and then generates a response to the query information. The generated response is then transmitted to the user support device 1.

[0022] The receiving unit 140 receives the response sent by the interactive response device 2. The received response is then output to the summarizing unit 150 and the presentation unit 160.

[0023] Summary Unit 150 generates a summary of the answer source based on the search performed by Answer Source Search Unit 120. Summary Unit 150 can generate the answer source summary using known methods such as graph-based, feature-based, or topic-based methods. Summary Unit 150 then outputs the generated summary to Presentation Unit 160.

[0024] The presentation unit 150 shows the user the answer received by the receiving unit 140 from the interactive answering device 2, and the summary generated by the summary unit 150. The presentation unit 160 can show the user the answer and the summary, for example, by displaying them side-by-side on the display device 70. In this case, the presentation unit 160 displays the answer and the summary in a distinguishable manner. For example, the answer and the summary can be displayed in different display fields or by changing the text colors, fonts, formats, sizes, etc.

[0025] Fig. Figure 4 is a screen view showing an example of an answer from the interactive answer device 2 and a summary shown by the presentation unit 160. In the example of Fig. 4. The user requests a procedure for rotating coordinates. In response, an answer from the interactive answer device 2 and a summary of instructions (an answer source) for coordinate rotation are displayed side by side. The user can refer to the summary of instructions as needed while referring to the answer from the interactive answer device 2. Even if there is an error in the answer from the interactive answer device 2, the user can easily check for errors by referring to the summary of instructions.

[0026] When the user queries the interactive response device 2, the user support device 1, according to this embodiment with the setup described above, displays a summary of an answer source corresponding to the answer of the interactive response device 2 alongside the answer itself. The user can view the summary of the answer source while referring to the answer of the interactive response device 2 and easily verify whether the answer is reliable. Therefore, the burden on the user when verifying the answer can be reduced. Generally, the interactive response device 2, which uses the LLM, provides an answer in a natural sentence; however, there may be an issue with accuracy.On the other hand, the summary of the answer source, such as instructions, appears in a somewhat unnatural sentence, but is correct in many cases within the context of the description. The user support device 1 according to this embodiment offers a user support function that takes advantage of this.

[0027] In a modified example of the user support device 1 according to this embodiment, the summary unit 150 can further select a sentence to be used in the summary from the answer source that resulted from the search. Fig. Figure 5 shows the functions of the summary unit 150 of the user support device 1 according to this modified example as a schematic block diagram. The summary unit 150 according to this modified example has a subdivision unit 152, a scoring calculation unit 154, and a selection unit 156.

[0028] Subdivision Unit 152 analyzes sentences from the answer source searched by Answer Source Search Unit 120 and subdivides the sentences into units of predefined sentences. For example, Subdivision Unit 152 can divide the answer source into units of predefined meaningful sentences such as chapters or sections. The subdivision might, for instance, divide a manual into sentences for individual technical units.

[0029] The scoring unit 154 calculates a score indicating the similarity between each sentence unit, subdivided by the subdivision unit 152, and the response from the interactive response device 2. This similarity score can be calculated, for example, based on the ratio, total number, etc., of the words in the response from the interactive response device 2 compared to the words in the subdivided sentence unit. This score can increase as the number of words in the response from the interactive response device 2 increases compared to the number of words in the sentence unit.

[0030] The selection unit 156, based on the score calculated by the scoring unit 154, selects a given sentence unit from several sentence units subdivided by the subdivision unit 152. The selection unit 156 can choose a sentence unit with the highest score. The summary unit 150 then generates a summary based on the sentence unit selected by the selection unit 156.

[0031] It should be noted that in this modified example, presentation unit 160 can change the display of the summary, if it is displayed, depending on the rating calculated by rating calculation unit 154. For example, the reliability of the summary can be expressed by highlighting characters differently depending on the rating level.

[0032] User Assistance Device 1, according to this modified example, selects a sentence unit from the response sources that is similar to the response from Interactive Response Device 1 and generates a summary based on that sentence unit. Therefore, it is easy for the user to accurately verify the reliability of the response by reviewing a summarized overview.

[0033] Although the embodiments of the disclosure have been described in detail above, the disclosure is not limited to the individual embodiments described above. Various additions, substitutions, modifications, partial omissions, etc., are possible in these embodiments without deviating from the core of the disclosure or from the idea and purpose of the disclosure derived from the content described in the claims and their equivalents. For example, in the embodiments described above, the sequence of individual activities and the sequence of individual processes are given as examples and are not limited to them. The same applies if numerical values ​​or mathematical expressions have been used in the description of the embodiments described above.

[0034] The following are supplementary remarks relating to the embodiments of the present disclosure. (Supplementary Note 1)

[0035] A user support device (1) according to an embodiment of the present disclosure comprises a query acquisition unit (100) configured to obtain information expressed in natural language that reflects the query content; a feature vector calculation unit (110) configured to calculate a feature vector for the query content; an answer source storage unit (200) configured to store at least one answer source for a query; an answer source search unit (120) configured to search for one or more answer sources from the answer source storage unit based on the feature vector; and a transmission unit (130) configured to transmit the query content and the answer source resulting from the search to an interactive response device.a receiving unit (140) configured to receive a response from the interactive response device; a summarizing unit (150) configured to generate a summary of the response source based on the response source obtained from the search; and a presentation unit (160) configured to output the received response and the summary. (Supplementary Note 2)

[0036] The summary unit (150) with which the user support device (1) is provided according to another embodiment of the present disclosure comprises a subdivision unit (150) configured to subdivide the answer source resulting from the search; a scoring calculation unit (154) configured to assign a score to a result of the subdivision using the answer obtained from the interactive answer device (2); and a selection unit (156) configured to make a selection based on the score. (Supplementary Note 3)

[0037] The response source storage unit (200), with which the user support device (1) is provided according to another embodiment of the present disclosure, stores a text of a certain length together with a feature vector for each text, and the response source search unit (120) performs the search on the basis of a distance between the feature vector of the query content and a feature vector of the response source. (Supplementary Note 4)

[0038] The presentation unit (160) with which the user support device (1) is provided according to another embodiment of the present disclosure outputs the received content and the content of the summary in a distinguishable manner. (Supplementary Note 5)

[0039] The presentation unit (160) with which the user support device (1) is provided according to another embodiment of the present disclosure modifies the output content according to the evaluation calculated by the summary unit (150). (Supplementary Note 6)

[0040] In a user support method according to another embodiment of the present disclosure, a computer performs a step to obtain information expressed in a natural language that represents the query content; a step to compute a feature vector for the query content; a step to search for at least one response source based on the feature vector from a response source storage unit (200) configured to store at least one response source for a query; a step to transmit the query content and the response source obtained from the search to an interactive response device (2); a step to receive a response from the interactive response device (2); a step to generate a summary of the response source based on the response source obtained from the search;and a step to output the received answer and the summary.; EXPLANATION OF REFERENCE SIGNS 1 User Support Device 2 INTERACTIVE ANSWER DEVICE 4 INDUSTRIAL MACHINE 5 NETWORK 6 FOG COMPUTERS 7 CLOUD SERVER 11 CPU 12 ROM 13 RAM 14 NON-FLAMMABLE STORAGE 15, 17, 18, 20 INTERFACE 22 BUS 70 DISPLAY DEVICE 71 INPUT DEVICE 72 EXTERNAL DEVICE 100 REQUEST RECEIPT UNIT 110 PERFORMANCE VECTOR CALCULATION UNIT 120 ANSWER SOURCE SEARCH UNIT 130 TRANSMISSION UNIT 140 RECEIVING UNIT 150 SUMMARY UNIT 152 SUBDIVISION UNIT 154 VALUATION UNIT 156 VOTING UNIT 160 PRESENTATION UNIT 200 ANSWER SOURCE STORAGE UNIT QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] JP 2022 - 112 541 A

[0002] JP 2019 - 160 286 A

[0002]

Claims

[1] User support device, comprising a query acquisition unit configured to obtain information expressed in a natural language that reflects the query content; a feature vector calculation unit that is set up to calculate a feature vector for the query content; a response source storage unit configured to store at least one response source for a query; an answer source search unit configured to search for one or more answer sources from the answer source storage unit based on the feature vector; a transmission unit configured to transmit the query content and the response source resulting from the search to an interactive response device; a receiving unit configured to receive a response from the interactive response device; a summary unit that is set up to generate a summary of the answer source based on the answer source that resulted from the search; and a presentation unit that is set up to output the received answer and the summary. [2] User support device according to claim 1, wherein the summary unit a subdivision unit that is set up to subdivide the answer source that resulted from the search; a scoring calculation unit configured to assign a score to a result of the subdivision using the response received from the interactive response device; and an electoral unit that is set up to conduct an election based on the evaluation, exhibits. [3] User support device according to claim 1, wherein The response source storage unit, together with a feature vector for each text, stores a text of a specific length, and The answer source search unit performs the search based on a distance between the feature vector of the query content and a feature vector of the answer source. [4] User support device according to claim 1, wherein the presentation unit outputs the received content and the content of the summary in a distinguishable manner. [5] User support device according to claim 2, wherein the presentation unit modifies the output content according to the rating calculated by the summary unit. [6] User support procedure performed by a computer, wherein the user support procedure comprises the steps to obtain information expressed in natural language that reflects the content of the request; to calculate a feature vector for the query content; to search for at least one answer source based on the feature vector from an answer source storage unit configured to store at least one answer source for a query; to transmit the query content and the answer source resulting from the search to an interactive response device; to receive a response from the interactive response device to generate a summary of the answer source based on the answer source that resulted from the search; and This includes displaying the received answer and the summary.

Images