User assistance device and computer-readable storage medium

Abstract

A user assistance device according to the present disclosure acquires a query relating to an object, classifies relevant information relating to the object into a subset on the basis of the query, generates a prompt for summarizing at least a portion of the subset, transmits the prompt to an interactive response device, and receives a response generated using the prompt.

Description

User assistance device and computer-readable storage medium

[0001] The present disclosure relates to a user assistance device and a computer-readable storage medium.

[0002] Currently, there are language generation models that generate natural language. The language generation model pre-learns text data and solves language processing tasks (such as document generation, fill-in-the-blank problems, machine translation, question and answer, etc.). The language generation model generates sentences based on the learned information. RAG (Retrieval-Augmented Generation) is a mechanism for supplementing knowledge to the language generation model. The language generation model improves the response accuracy by combining knowledge obtained from the outside. For example, there is Patent Document 1.

[0003] In the language generation model, output text is generated from input text. However, since there is a limit to the length of the text that can be input, it is difficult to input all relevant information when there is a lot of it. Also, since the processing time and cost (price) increase depending on the length of the input text, it is desirable that the text length for one inquiry be as short as possible. Also, in RAG, due to its characteristics, cases with a large number of cases may be prioritized. In that case, a response is generated based on partial information rather than a response based on appropriate information.

[0004] Japanese Patent No. 7586541

[0005] It is desired to improve the accuracy of the response of the generation model.

[0006] The user assistance device according to the present disclosure includes a query acquisition unit that acquires a query regarding a target, a related information classification unit that generates classification conditions based on the query and classifies related information regarding the target into subsets, a prompt generation unit that generates a prompt for summarizing at least a part of the subsets, a transmission unit that transmits the prompt to an interactive response device, and a reception unit that receives a response generated using the prompt from the interactive response device.

[0007] This is a block diagram of the user assistance device of the first embodiment. This is a diagram showing an example of the input screen for the maintenance log. This is a diagram showing an example of the maintenance log. This is a diagram showing an example of the maintenance log after filtering. This is a schematic diagram showing the operation of the user assistance device. This is a diagram showing an example of the display screen for the aggregated results. This is a diagram showing an example of the display screen for the aggregated results. This is a block diagram of the user assistance device of the second embodiment. This is a diagram showing an example of the display screen for the aggregated results with statistics added. This is a block diagram of the user assistance device of the third embodiment. This is a hardware configuration diagram of the user assistance device.

[0008] Embodiments of this disclosure will be described below with reference to the drawings. In the following description, components having the same or similar functions will be denoted by the same reference numerals. Duplication of these components may be omitted.

[0009] In this application, "based on XX" means "based on at least XX," and includes cases where it is based on another element in addition to XX. Furthermore, "based on XX" is not limited to cases where XX is used directly, but also includes cases where it is based on something that has been calculated or processed. "XX" is any element (for example, any information).

[0010] The assistive device 100 will now be described. The assistive device 100 is an information processing device such as a server. The assistive device 100 acquires queries related to a target. The target may include industrial machinery, home appliances, personal computers (PCs), etc. The target is not particularly limited. The operation history of other machines / devices may also be used as the target. A user operating the target machine / device outputs a query to the assistive device. The assistive device responds to the query based on the operation history of the machine / device.

[0011] The response is primarily based on a summary of the operation history. The assistive device 100 classifies the operation history into subsets and summarizes the subsets. The assistive device 100 extracts commonalities within the subsets, extracts important information within the subsets, and extracts information highly relevant to the query. In this case, query information may be added to the prompt.

[0012] The user assistance device 100 causes the interactive response device to summarize the subset. During the summarization process, a representative of the subset may be selected and transmitted to the interactive response device.

[0013] The user assistance device 100 may include an interactive response device. That is, the interactive response device provided in the user assistance device 100 may perform summarization / aggregation.

[0014] The prompt may include instruction information (the content of the task to be performed by the LLM), constraints, the output format, and a sample of the output format. Furthermore, a generative model may be used to process the summary result. For example, the summary result may be processed using a prompt that includes instructions to simplify it. In this case, the processed result may be used in a subsequent aggregation process. The term "aggregation" has two meanings: one is to gather many things together, and the other is to organize and summarize them.

[0015] The generative models used for "summarization" and "aggregation" may be the same model or different models.

[0016] In this embodiment, the user assistance device 100 will be described in relation to the maintenance log (operation history) of industrial machinery. The user assistance device 100 is connected to the terminal of the industrial machinery's maintenance personnel. Maintenance personnel refer to users who have the authority to maintain the industrial machinery. The user assistance device 100 responds to queries received from maintenance personnel. The user assistance device 100 may be part of an external network such as a cloud, or part of the factory's internal network. The terminal may be a mobile phone, personal computer, tablet, etc.

[0017] [First Embodiment] Figure 1 is a block diagram of the user assistance device 100 according to the first embodiment. The user assistance device 100 comprises a related information storage unit 11, a query acquisition unit 12, a related information classification unit 13, a related information retrieval unit 14, a prompt generation unit 15, a transmission unit 16, a reception unit 17, and an aggregation unit 18.

[0018] The related information storage unit 11 stores information related to the maintenance of industrial machinery. This maintenance-related information includes, for example, the maintenance log (operation history) of the industrial machinery. The maintenance log is a record of maintenance, servicing, inspection, and repairs previously performed by the industrial machinery's maintenance staff. The maintenance log may also include audio, photographs, and videos.

[0019] Figure 2 shows an example of a maintenance log input screen. Maintenance personnel record the necessary information according to this input screen. Note that maintenance logs can also be recorded by reading data from paper documents using character recognition or similar methods.

[0020] Maintenance logs consist of specific items. These items include the name of the industrial machine, the name of the control device, the date and time of maintenance, the alarm number, replacement parts, the nature of the failure, and the details of the action taken. The information contained in these items may be extracted from free text through preprocessing or obtained from external sources. The items described above are examples of maintenance log items and are not exhaustive.

[0021] The related information storage unit 11 stores the maintenance logs collected so far. Figure 3 shows an example of a maintenance log stored in the related information storage unit 11. The maintenance log in Figure 3 is a CSV file with data separated by commas, but the file format of the maintenance log is not particularly limited. A maintenance log consists of multiple items. If an item is a categorical variable, the maintenance log can be classified by the content of the item. When an item is a categorical variable, the content of the item is called an attribute.

[0022] For example, in the example in Figure 3, the attribute of the item "Industrial Machinery" is "MCH-2000". The attribute of the item "Control Device" is "MODEL A", the attribute of the item "Date and Time" is "2024 / 5 / 4", and the attribute of the item "Alarm Number" is "ALM159".

[0023] The fields "Details of the malfunction" and "Details of the action taken" are free text. If the data included in the fields is in the format of free text, audio, or image, the saved logs can be classified using natural language processing, speech recognition, or image processing.

[0024] The query acquisition unit 12 acquires queries from the industrial machinery maintenance personnel. Here, maintenance personnel refer to users who have the authority to maintain the industrial machinery.

[0025] The related information classification unit 13 determines the classification conditions for maintenance logs based on the query. Specifically, it extracts information from the query and determines the conditions (called classification conditions) for filtering, classifying, and clustering the maintenance logs based on the extracted information.

[0026] Existing language processing techniques are used to extract information from queries. For example, if the query is the text "ALM95 communication error occurred," the alarm number "ALM95" can be extracted from the query. The alarm number can be extracted, for example, by searching for the string "ALM" + "number." In addition, language processing techniques such as character recognition can be used to extract alarm-related words such as "communication error" and "error occurred" from the text.

[0027] The information extracted from the query becomes the classification criteria for the maintenance log. Furthermore, the attributes of the maintenance log related to the information extracted from the query also become classification criteria for the maintenance log. Additionally, the free text contained in the maintenance log can be clustered to classify the log.

[0028] Here, we will explain the term filtering. Filtering extracts only a subset that meets certain conditions. In this embodiment, filtering is also included in classification. If a query contains a specific keyword, maintenance logs related to that keyword are classified. For example, if the query is "Tell me about amplifier failure countermeasures," the keywords "amplifier" and "failure countermeasures" will be extracted, but if the word "amplifier" is registered as a keyword, maintenance logs related to "amplifier" will be classified. Keywords may be registered in advance, or users may add / delete them as they wish.

[0029] The saved logs can also be classified by the attributes of the items. In this embodiment, for convenience, items that serve as filtering conditions are called required items, and items that serve as classification conditions are called optional items. The combination of required and optional items may be predetermined or specified by the user. For example, "Alarm Number" may be a required item and "Replacement Part" may be an optional item.

[0030] Once the classification conditions are set in this way, the related information classification unit 13 recognizes the word "ALM95" extracted from the query as an alarm number and filters the maintenance logs by alarm number "ALM95". Through this filtering, maintenance logs other than those with alarm number "ALM95" are excluded. Furthermore, by classifying the filtered maintenance logs by the attribute of the optional item "replacement parts", the maintenance logs can be classified.

[0031] Here is an example of an optional item. Optional items are those related to maintenance work. Typically, maintenance work can be classified into replacement, verification, setting, correction, and adjustment. By designating items related to each maintenance work (e.g., "replacement parts," "verification points," "setting points," "correction points," "adjustment points") as optional items, maintenance logs can be classified by maintenance work. For example, let's say the optional item is "replacement parts." The maintenance log is then classified by the attribute of "replacement parts." If the "servo amplifier" and "I / O module" have been replaced in the maintenance log filtered by "ALM95," then the log will be classified as a maintenance log with the attributes "servo amplifier" and "I / O module."

[0032] Multiple optional fields may be set. For example, if "Modification Location" is an optional field, it may include attributes such as "Parameter No. 003". In this way, maintenance logs can be classified using attributes as categories. Classification can also be hierarchical. For example, maintenance work can be initially classified into "Replacement," "Confirmation," "Setting," "Modification," and "Adjustment," and then further classified by "Replacement Parts." Required and optional fields may be made selectable by the user.

[0033] Next, we will explain the classification of free text. Free text (documents) cannot be used directly for classifying maintenance logs. Therefore, the related information classification unit 13 uses clustering to classify the maintenance logs into subsets.

[0034] Here are two examples of clustering. The first uses word distribution. In this method, free text is broken down into words, and the text is classified based on the distribution of the words that appear. Specifically, this includes topic models. Topic models are models based on the assumption that the words contained in a text indicate the subject (topic) of the text. Specifically, these include, but are not limited to, LSI (Latent Semantic Index), PLSI (Probabilistic Latent Semantic Indexing), and LDA (Latent Dirichlet Allocation).

[0035] The second method is mathematical clustering. The related information classification unit 13 converts the free text into some kind of feature. Generally, the feature can be hundreds of items with hundreds of dimensions. By converting the free text into feature, the text can be treated as numerical values. The related information classification unit 13 clusters (classifies) the feature using statistical operations. Clustering methods include, but are not limited to, k-nearest neighbors, Gaussian mixture models, and DBSCAN (Density-based spatial clustering of applications with noise).

[0036] Figure 4 shows a specific example. Figure 4 shows an example of maintenance logs when filtered by a fictitious alarm "ALM09". As a premise, "ALM09" is an alarm related to a fan. Assume that seven maintenance logs are extracted as a result of the filtering. Assume that the content of the "Processing" item in the maintenance log is free text as follows: Maintenance log 1 is "The cooling fan had stopped, so it was converted", Maintenance log 2 is "The cooling fan stopped working, so it was replaced", Maintenance log 3 is "The cooling fan was not working, so it was replaced", Maintenance log 4 is "The cooling fan was not working, so it was replaced", Maintenance log 5 is "The load status was changed", Maintenance log 6 is "The load was adjusted", Maintenance log 7 is "Parameter (No. 3382) was modified".

[0037] Ideally, clustering the seven maintenance logs 1-7 would result in three subsets. Subset 1 (maintenance logs 1-4) relates to fan replacement, Subset 2 (maintenance logs 5-6) relates to load, and Subset 3 (maintenance log 7) relates to parameters. In this way, maintenance logs can be classified using free text.

[0038] The related information retrieval unit 14 selects a maintenance log. The selected maintenance log is sent to the interactive response device as related information. The interactive response device includes, for example, a generative model such as a large-scale language model. The generative model may generate not only text, but also audio, images, etc. The processing of the related information retrieval unit 14 is divided into two parts. The first part is the process of selecting a representative (maintenance log) of a subset. The second part is the process of selecting a subset. The two parts can be performed in any order.

[0039] In the first step, a portion of the maintenance logs included in the subset are selected as representatives. Selecting representatives reduces the number of maintenance logs sent to the interactive response device. A single subset may contain tens or hundreds of maintenance logs. Sending all maintenance logs is costly, so costs are reduced by sending only one or a small number of maintenance logs. There are no particular limitations on the method of selecting representatives. They may be selected randomly. Alternatively, the stored logs may be scored, and selection may be based on the scores. An example of a score for stored logs is text length. Since longer texts contain more information, logs with longer texts may be selected. Alternatively, scores calculated using a deep learning model may be used. For example, whether the log includes the steps taken to investigate the cause of a malfunction may be used as the score. Another option is to vectorize the maintenance logs included in the subset and select maintenance logs with large vector distances. By selecting maintenance logs that are far apart, a variety of information can be obtained. An upper limit on the number of maintenance logs to select may also be set. Instead of selecting representatives of the maintenance logs, the entire set may be selected.

[0040] In the second process, a subset to be transmitted to the interactive response device is selected. As a method for selecting the subset, for example, a subset with high relevance to the query is selected. To select a subset with high relevance to the query, the security logs and the query included in the subset are vectorized, and the relevance can be calculated by calculating the distance between the vectors. Also, the relevance can be calculated from the word distribution between the security logs and the query included in the subset. Subsets with low relevance to the query may be excluded from the candidates to be transmitted to the interactive response device.

[0041] There is also a method of selecting a subset based on the number of security logs included in the subset. Since a subset with a small number of security logs is a rare case, it can be determined that it should be excluded. Conversely, it can be determined that subsets with a large number of security logs should also be excluded because they contain common content. The conditions for selecting the subset may be specified by the user.

[0042] The prompt generation unit 15 generates a prompt based on the security logs selected by the related information search unit 14. The prompt is composed of, for example, a command and a security log. The command is, for example, the text "Please summarize the following security logs". Here, a prompt for summarizing multiple classifications together may be created. The prompt generation unit 15 generates the above prompts for each classification of the security logs.

[0043] The transmission unit 16 transmits the prompt generated by the prompt generation unit 15 to the interactive response device. The interactive response device summarizes the security logs for each classification of the security logs according to the prompt. The interactive response device may generate not only text but also voice, images, etc. as the summary. The reception unit 17 receives the summary result of the interactive response device.

[0044] The aggregation unit 18 aggregates the summary results to create a response text. The aggregation unit 18 outputs the response text to the terminal of the security guard.

[0045] Figure 5 shows the operation of the user assistance device 100. The related information storage unit 11 stores maintenance logs as information related to the maintenance of industrial machinery. Maintenance logs are records of past maintenance. The query acquisition unit 12 acquires queries from the maintenance worker's terminal. The related information classification unit 13 generates classification conditions based on the queries. Classification conditions include keywords, required items, optional items, and clustering conditions. Required and optional items do not necessarily have to be set. Keywords are extracted from the query. Required items are attributes that the maintenance log must contain. In the example above, "alarm number" is a required item. If the query contains "alarm number", the related information classification unit 13 filters the maintenance logs by "alarm number". Optional items are items used to classify the maintenance logs. The related information classification unit 13 classifies the maintenance logs according to the attributes of the optional items. This generates a subset of the maintenance logs.

[0046] The related information classification unit 13 clusters data other than categorical variables such as free text, images, and audio using means such as text classification, image classification, and audio classification.

[0047] The related information retrieval unit 14 selects a subset to send to the interactive response device and selects a maintenance log that represents the subset. The prompt generation unit 15 generates a prompt to query a summary of the selected maintenance log. The transmission unit 16 sends the prompt to the interactive response device, and the reception unit 17 receives the response from the interactive response device. The aggregation unit aggregates the responses from the interactive response device and outputs them to the maintenance worker's terminal.

[0048] Figures 6 and 7 are examples of the aggregation results. The aggregation result in Figure 6 is displayed in a list format. In the aggregation result of Figure 6, the security guard's query "ALM195 communication abnormality occurred" is displayed in "<Query>", and subsequently, in "<Answer>", "Replacement part: servo amplifier", "If an error occurs when power is turned on, replace the servo amplifier", "If alarm PC51 also occurs, replace the servo amplifier", "Replacement part: IO module", "Check the fuse of the I / O unit and replace it if necessary", "Check the poor contact of the connector of the I / O unit and replace it if necessary",... and the maintenance procedures are displayed in a list format. The aggregation result of Figure 7 is in a tree format. In the aggregation result of Figure 7, the security guard's query "ALM195 communication abnormality occurred" is displayed in "<Query>", and subsequently, "<Answer>" is displayed in a tree format.

[0049] As described above, the user assistance device classifies the maintenance logs into subsets based on the queries input by the security guard, selects a part of the classified maintenance logs, and aggregates the content of the subsets. Therefore, findings can be obtained from the set of similar maintenance logs and the accuracy is improved. Also, since the maintenance logs are filtered, classified, and selected, the text length to be transmitted to the interactive response device is shortened, and the costs of time and price are reduced.

[0050] Since the keyword, classification item, and clustering method can be selected by the user, the user can obtain the target information while trying and making mistakes.

[0051] By extracting keywords from the query and classifying the maintenance logs, a response can be generated from the maintenance logs that match the query.

[0052] There are mandatory items and optional items for the classification items. Since the attributes of the mandatory items are mandatory, only the maintenance logs with specific attributes are filtered. By filtering the maintenance logs, unnecessary information can be excluded. As a result, the text length of the maintenance logs can be suppressed.

[0053] By selecting and summarizing the maintenance logs highly relevant to the query, the response accuracy of the query is improved, and since unnecessary maintenance logs are excluded, the text length can be suppressed.

[0054] [Second Embodiment] Figure 8 is a block diagram of the user assistance device 100 of the second embodiment. The same configuration as the user assistance device of the first embodiment will not be explained. In the second embodiment, the related information classification unit 13 calculates statistical quantities of the classification results. The statistical quantity is the number of elements in the maintenance log. The number of elements in the maintenance log is the number of maintenance logs included in each classification. In addition, the number of elements classified by season from the date can be calculated. Furthermore, when clustering is performed using free text, the top-level texts of topics can also be obtained as statistical quantities.

[0055] The related information classification unit 13 outputs the calculated statistics to the aggregation unit. The aggregation unit 18 includes the input statistics in the aggregation result. Figure 9 shows the aggregation result including the statistics. In Figure 9, the first line of "<Answer>" displays the statistics "15 servo amplifiers (10 of which were in summer)", and the fourth line displays the statistics "10 I / O modules (10 of which were in mist environments)".

[0056] Statistical data allows caregivers to determine the circumstances and frequency under which events occurred.

[0057] [Third Embodiment] Figure 10 is a block diagram of the user assistance device 100 of the third embodiment. The same configuration as the user assistance device 100 of the first embodiment will not be described. The aggregation unit 18 of the third embodiment obtains summaries of individual subsets and queries the interactive response device to generate a response sentence by aggregating them. The prompt may include instructions to summarize individual subsets and instructions to aggregate the summaries. The format of the aggregation may also be specified using the prompt. The prompt may also include instructions to process the summary results. For example, the prompt may include an instruction to simplify the summary results. The aggregation unit obtains the aggregation results from the interactive response device and outputs them to the terminal.

[0058] The hardware configuration of the user assistance device 100 to which this disclosure is applied will be described below. Figure 11 is a hardware configuration diagram of the user assistance device 100. As an example, the user assistance device 100 is implemented as a computer comprising a processor 101, main memory 102, auxiliary memory 103, network interface 104, and device interface 105, which are connected via a bus 106.

[0059] The user assistance device 100 in Figure 8 has one of each component, but it may have multiple identical components. Also, although Figure 8 shows one user assistance device 100, the software may be installed on multiple computers, and each of these computers may execute the same or different parts of the software's processing. In this case, it may be a distributed computing configuration in which each computer communicates via a network interface to execute processing.

[0060] The various processes in the embodiments described above may be executed in parallel using one or more processors 101, or using multiple computers via a network. Alternatively, the various processes may be distributed to multiple arithmetic cores in the processor 101 and executed in parallel. The processor 101 may be an electronic circuit (processing circuit, processing circuit, processing circuit, CPU, GPU, FPGA, ASIC, etc.) that performs at least one of computer control or calculation. Furthermore, the processor 101 may be a general-purpose processor, a dedicated processing circuit designed to perform a specific calculation, or a semiconductor device that includes both a general-purpose processor and a dedicated processing circuit.

[0061] The main memory 102 may store instructions executed by the processor 101 and various data, and the information stored in the main memory 102 may be read by the processor 101. The auxiliary memory 103 is a memory device other than the main memory 102.

[0062] The network interface 104 is an interface for connecting to a communication network by wire or wireless connection. The communication network may be a WAN (Wide Area Network), LAN (Local Area Network), PAN (Personal Area Network), or a combination thereof. The user assistance device 100 exchanges information with the external device 107 via the network interface 104. The device interface 105 is an interface such as USB for direct connection to the external device 108. The external device 107 is a device connected to the user assistance device 100 via a network. The external device 108 is a device directly connected to the user assistance device 100. The external devices 107 and 108 may be personal computers, tablet terminals, smartphones, or other devices. The external devices may also be storage devices.

[0063] While embodiments of this disclosure have been described in detail above, this disclosure is not limited to the individual embodiments described above. These embodiments can be added, replaced, modified, partially deleted, etc., in any way that does not depart from the spirit of the invention or from the idea and intent of this disclosure derived from the claims and their equivalents. For example, the order of operations and processes in the embodiments described above are shown as examples only and are not limited thereto. The same applies when numerical values ​​or mathematical formulas are used in the description of the embodiments described above.

[0064] The following are annotations relating to embodiments of the present disclosure. (Annotation 1) The user assistance device (100) includes a query acquisition unit (12) that acquires queries relating to an object, a related information classification unit (13) that classifies related information relating to the object into subsets based on the queries, a prompt generation unit (15) that generates prompts to summarize at least a part of the subsets, a transmission unit (16) that transmits the prompts to an interactive response device, and a receiving unit (17) that receives responses generated using the prompts from the interactive response device. (Annotation 2) The user assistance device (100) includes a related information retrieval unit (14) that selects a representative of the subsets. (Annotation 3) The receiving unit (17) receives the summaries of two or more subsets from the interactive response device. (Annotation 4) The user assistance device (100) includes an aggregation unit (18) that aggregates the summaries of two or more subsets. (Note 5) The aggregation unit (18) sends a prompt to the interactive response device to aggregate the summaries of two or more subsets, and receives the aggregation results from the interactive response device. (Note 6) The related information classification unit (13) extracts words from the query, and if the word is a keyword, it classifies related information related to the keyword. (Note 7) The related information consists of multiple items, and the related information classification unit (13) classifies the related information according to the attributes of the items. (Note 8) The related information classification unit (13) filters the related information based on the attributes of the items. (Note 9) The related information includes free text, and the related information classification unit (13) classifies the free text using clustering. (Note 10) The clustering classifies the related information based on the distribution of words contained in the free text. (Note 11) The clustering converts the free text into features and mathematically classifies the related information. (Note 12) The related information classification unit (13) generates classification conditions for related information relating to the subject based on the query. (Note 13) The related information retrieval unit (14) uses the relationship between the related information included in the subset and the query as the selection condition for the subset.(Note 14) The related information retrieval unit (14) uses the number of related pieces of information included in the subset as a selection criterion for the subset. (Note 15) The related information retrieval unit (14) uses the magnitude of the distance between related pieces of information included in the subset as a selection criterion for the related information. (Note 16) The related information retrieval unit (14) uses the amount of information in the related information as a selection criterion for the related information. (Note 17) The related information classification unit (13) calculates a statistical quantity for each subset, and the aggregation unit (18) includes the statistical quantity in the display screen along with the aggregation result. (Note 18) Computer-readable storage media (102, 103, 107, 108) record a program that causes the computer to operate as a query acquisition unit (12) that acquires queries about an object, a related information classification unit (13) that classifies related information about the object into subsets based on the queries, a prompt generation unit (15) that generates prompts to summarize at least a part of the subsets, a transmission unit (16) that transmits the prompts to an interactive response device, and a reception unit (17) that receives responses generated using the prompts from the interactive response device.

[0065] 100 User assistance device 11 Related information storage unit 12 Query acquisition unit 13 Related information classification unit 14 Related information retrieval unit 15 Prompt generation unit 16 Transmission unit 17 Receiving unit 18 Aggregation unit 101 Processor 102 Main memory 103 Auxiliary memory

Claims

1. A user assistance device comprising: a query acquisition unit that acquires queries concerning an object; a related information classification unit that classifies related information concerning the object into subsets based on the queries; a prompt generation unit that generates a prompt that summarizes at least a part of the subsets; a transmission unit that transmits the prompt to an interactive response device; and a receiving unit that receives a response generated using the prompt from the interactive response device.

2. The user assistance device according to claim 1, further comprising a related information retrieval unit for selecting a representative of the subset.

3. The user assistance device according to claim 1, wherein the receiving unit receives the summary results of two or more subsets from the interactive response device.

4. The user assistance device according to claim 3, further comprising an aggregation unit for aggregating the summarization results of the two or more subsets.

5. The user assistance device according to claim 4, wherein the aggregation unit transmits a prompt to the interactive response device to aggregate the summaries of two or more subsets, and receives the aggregation results from the interactive response device.

6. The user assistance device according to claim 1, wherein the related information classification unit extracts words from the query, and if the word is a keyword, classifies related information related to the keyword.

7. The user assistance device according to claim 1, wherein the related information consists of multiple items, and the related information classification unit classifies the related information according to the attributes of the items.

8. The user assistance device according to claim 7, wherein the related information classification unit filters the related information based on the attributes of the items.

9. The user assistance device according to claim 1, wherein the related information includes free text, and the related information classification unit classifies the free text using clustering.

10. The user assistance device according to claim 9, wherein the clustering classifies related information based on the distribution of words contained in the free text.

11. The user assistance device according to claim 9, wherein the clustering converts the free text into features and mathematically classifies related information.

12. The user assistance device according to claim 1, wherein the related information classification unit generates classification conditions for related information relating to the subject based on the query.

13. The user assistance device according to claim 2, wherein the related information retrieval unit uses the relationship between the related information included in the subset and the query as the selection condition for the subset.

14. The user assistance device according to claim 2, wherein the related information retrieval unit uses the number of related information items included in the subset as the selection condition for the subset.

15. The user assistance device according to claim 2, wherein the related information retrieval unit uses the magnitude of the distance between related information included in the subset as a selection criterion for the related information.

16. The user assistance device according to claim 2, wherein the related information retrieval unit uses the amount of information of the related information as a selection criterion for the related information.

17. The user assistance device according to claim 4, wherein the related information classification unit calculates statistics for each subset, and the aggregation unit includes the statistics on the display screen along with the aggregation results.

18. A computer-readable storage medium recording a program that causes the computer to operate as: a query acquisition unit that acquires queries about an object; an related information classification unit that classifies related information about the object into subsets based on the queries; a prompt generation unit that generates prompts that summarize at least a part of the subsets; a transmission unit that transmits the prompts to an interactive response device; and a receiving unit that receives responses generated using the prompts from the interactive response device.

Images