CN115203492A - Query feedback device and method - Google Patents

Abstract

The invention provides a query feedback device, which comprises a transceiver circuit, a memory and a processor. The receiving and transmitting circuit is used for receiving input information; the memory is used for storing a plurality of instructions and a plurality of context databases; the processor is connected with the transceiver circuit and the memory and is used for executing a plurality of instructions: according to the input information, a plurality of situation intents respectively corresponding to the plurality of situation databases and a plurality of sample elements, identifying one of the situation databases, at least one input element and the input intention corresponding to the input information; and generating feedback information corresponding to the at least one input element according to the corresponding one of the situation databases, so as to output the feedback information through the transceiver circuit. In addition, a query feedback method is also disclosed.

Description

Query feedback device and method

Technical Field

The invention relates to a query feedback device and a method.

Background

An existing Office Automation (OA) system provides business functions of accessing, querying and managing internal data of a company for employees, thereby serving as an auxiliary tool for improving the work efficiency of the employees. However, when the employee needs to query the required data by using the OA system, the OA system may not query the required data of the employee, and the data queried by the OA system is not necessarily the required data of the employee. In addition, the integrity of the data queried by OA systems is often more inadequate. Therefore, how to accurately query the data required by the complete employee is a problem to be solved by those skilled in the art.

Disclosure of Invention

The invention provides a query feedback device, which comprises a transceiver circuit, a memory and a processor. The receiving and transmitting circuit is used for receiving input information; the memory is used for storing a plurality of instructions and a plurality of context databases, wherein the plurality of context databases respectively correspond to a plurality of different context intents, and the plurality of context databases are provided with a plurality of sample elements; the processor is connected with the transceiver circuit and the memory and is used for executing a plurality of instructions: according to the input information, the plurality of situation intents and the plurality of sample elements, identifying one of the situation databases, at least one input element and the input intents corresponding to the input information; and generating feedback information corresponding to the at least one input element according to one of the corresponding context databases to output the feedback information through the transceiver circuit, wherein the one of the context databases corresponds to the input intention.

The invention provides a query feedback method. The method comprises the following steps: receiving input information through a transceiver circuit; performing machine learning calculation on a plurality of situation intents or a plurality of situation intents and a plurality of sample elements through a processor to train a recognition model, wherein the plurality of sample elements are stored in a plurality of situation databases in a memory, and the plurality of situation databases respectively correspond to the plurality of situation intents; receiving, by a processor, input information, the processor performing recognizing at least one input element corresponding to the input information and an input intention using a recognition model; generating, by a processor, feedback information corresponding to at least one input element according to one of context databases, wherein the one of context databases corresponds to an input intent; and outputting the feedback information through the transceiver circuit.

Based on the above, the query feedback device and method of the present invention provide a flexible architecture, and can determine the input intention and the input elements according to the characters, images or voices inputted by the user. Thereby, the query feedback device may search various elements similar to the input element from a database corresponding to the input intention to generate feedback information. Therefore, the structure of the global conversation can lead the user to save the time for inquiring the information, can prevent the information drop caused by the insufficiency of the information of the user, or can lead the user to take the precautionary measures.

Drawings

FIG. 1 shows a block diagram of a query feedback device according to some embodiments of the present invention.

FIG. 2 illustrates a flow diagram of a query feedback method of some embodiments of the invention.

FIG. 3 shows a schematic diagram of a query feedback device of further embodiments.

Fig. 4 shows a flowchart of a query feedback method of another embodiment.

Description of reference numerals:

100: query feedback device

110: transceiver circuit

120: memory device

130: processor with a memory for storing a plurality of data

1201: dialogue processing module

1201 (1): intention recognition module

1201 (2): dialogue memory module

1201 (3): data retrieval module

1201 (4): dialogue generating module

1203: data processing module

1203 (1): database set

1203 (2): data collection module

1205: user report receiving module

200: user device

300: external network

S201 to S207, S401 to S405: step (ii) of

Detailed Description

Referring to fig. 1, the query feedback device 100 may include a transceiver circuit 110, a memory 120, and a processor 130. In practical applications, the query feedback device 100 is various electronic devices that can receive input signals from users/other devices, perform corresponding data search (e.g., web search, database search) according to the input signals, and provide corresponding responses, in one embodiment, the query feedback device 100 may be disposed in public spaces such as companies, offices, stations, etc. for different users to freely query information, in another embodiment, the query feedback device 100 may also be disposed in a private mobile phone or a computer owned by a specific user.

The transceiver circuit 110 may transmit and receive an input signal (the input signal may include a sound signal, a text signal or an image signal) wirelessly or by wire, and may perform operations such as low noise amplification, impedance matching, frequency mixing, up or down frequency conversion, filtering, amplification and the like on the input signal. The transceiver circuit 110 may accept input information from a user, wherein the input information may include various types of information such as text (e.g., numerical values, sentences or words), images, voice, or a combination thereof (e.g., the user may input text information using a physical keyboard of the user device to transmit the text information to the transceiver circuit 110). Transceiver circuitry 110 may also perform operations such as low noise amplification, impedance matching, mixing, up or down frequency conversion, filtering, amplification, and the like on the input signal.

The memory 120 may store a plurality of instructions and a plurality of context databases, wherein the context databases respectively correspond to a plurality of different context intents and the context databases have a plurality of sample elements. The processor 130 may be connected to the transceiver circuit 110 and the memory 120, and may load and execute the instructions.

In some embodiments, the plurality of context databases may respectively store a plurality of sample elements and a plurality of sentence elements corresponding to different context intentions, wherein a context intention indicates a user intention related to the corresponding context database, a sample element may be various words related to the corresponding context intention (for example, in case of context data related to epidemic diseases, words such as a region, a country name, a symptom, or a date), and a sentence element may be various sentences related to the corresponding context intention (for example, a description related to influenza).

For example, taking medical-related databases as an example, the memory 120 may store a plurality of contextual databases, wherein the contextual databases include a medical information database, a travel warning database, a rumor identification database, and an epidemic situation distribution prediction database. The databases may correspond to a plurality of contextual intents, wherein the contextual intents may include medical queries, travel queries, rumor queries, and epidemic queries. In other words, the medical information database, the travel warning database, the rumor identification database, and the epidemic situation distribution prediction database may correspond to the medical query, the travel query, the rumor query, and the epidemic situation query, respectively.

In detail, the medical information database may store various information related to medical treatment, the travel alert database may store various information related to travel alert and disease distribution, the rumor identification database may store various information related to known rumor information, and the epidemic situation distribution prediction database may store various information related to epidemic disease propagation prediction (for example, a linear regression algorithm is performed in advance using open data such as a global confirmed population statistical table, a global flight number statistical table, an outbound statistical table, and related news to generate various information related to epidemic disease propagation prediction).

Further, the medical information database may further store various sample elements related to medical query (e.g., illness, hospital or doctor, etc.), the travel alert database may further store various sample elements related to travel query (e.g., XX country, geological disaster or weather, etc.), the rumor identification database may further store various sample elements related to rumor query (e.g., tax, law violation or criminal liability, etc.), and the epidemic distribution prediction database may further store various sample elements related to epidemic query (e.g., new coronavirus (corona virus), influenza or epidemic area, etc.).

In another example, taking a plurality of databases related to enterprise internal information retrieval as an example, the memory 120 may store a plurality of context databases, wherein the context databases include a factory information database and a attendance query database. The databases may correspond to a plurality of contextual intents, wherein the contextual intents may include factory queries and attendance queries. In other words, the factory information database and the attendance query database may correspond to factory queries and attendance queries, respectively.

The factory information database may store various information related to companies, factories, and the like (e.g., various administrative data, wherein the administrative data may include parking information or meeting room locations, for example), and the information may be a structured data (e.g., in the form of a table), and the attendance query database may store various information related to attendance status of employees. Generally, when structured data are stored in the factory information database and the attendance query database, it is more beneficial to perform data query or data similarity comparison. The factory information database may also store various sample elements related to factory inquiries (e.g., conference room name, conference room location, extension number or area where the conference room is located, etc.), and the attendance inquiry database may also store various sample elements related to attendance inquiries (e.g., employee number or attendance-related information, etc.).

In another example, the memory 120 may store a plurality of context databases, for example, a plurality of databases associated with the plant, wherein the context databases include a plant information database and a machine status database. The databases may correspond to a plurality of contextual intents, which may include plant information queries and tool health status queries. The factory information database and the machine state database may correspond to a factory information query and a machine health status query, respectively.

The plant information database may store various information related to machines or devices in the plant (e.g., information related to various types of machines, legal personnel access management, attendance monitoring, material management, etc., and such information may be a data structure in which a question sentence corresponds to an answer sentence), and the machine status database may store various information related to various measurement data of the machines (e.g., an operation status, a routine maintenance status, or various data detections diagnosed when the machines are in operation, etc., wherein such data may be detected waveform diagrams).

The plant information database may store various sample elements related to a plant information query (e.g., a model number of a tool or a location of a plant, etc.), and the tool status database may store various sample elements related to a tool health status query (e.g., a type of tool or a type of measurement data, etc.).

In some embodiments, the transceiver circuit 110 is, for example, one or a combination of a transmitter circuit, an analog-to-digital converter, a digital-to-analog converter, a low noise amplifier, a mixer, a filter, an impedance matcher, a transmission line, a power amplifier, one or more antenna circuits, and a local storage medium element. In some embodiments, the memory 120 may be any form of fixed or removable memory, hard disk, or the like, or any combination thereof. In some embodiments, the processor 130 is, for example, a Central Processing Unit (CPU), or other programmable general purpose or special purpose Micro Control Unit (MCU), a microprocessor (microprocessor), a Digital Signal Processor (DSP), a programmable controller, an Application Specific Integrated Circuit (ASIC), or other similar components or combinations thereof. In some embodiments, the processor 130 may connect the transceiver circuit 110 and the memory 120 in a wired or wireless manner.

The following will refer to fig. 1 to fig. 3 to explain the detailed steps of the query feedback method shown in fig. 2 by referring to the operation relationship between the elements in the query feedback device 100. The method of the embodiment shown in fig. 2 is suitable for the query feedback device 100 of fig. 1 and fig. 3, but is not limited thereto.

First, in step S201, the transceiver circuit 110 receives input information. In other words, the user can receive various types of input information by querying the transceiver circuit 110 in the feedback apparatus 100. In some embodiments, the user may input one of text information, image information, voice information, or a combination of the above types of input information by using a physical or virtual keyboard, mouse, microphone, or touch panel of the emissary device 200 (e.g., a smart phone or a notebook computer) to transmit the input information to the transceiver circuit 110.

In some embodiments, the processor 130 may convert various types of input information (i.e., the audio signal, the text signal or the video signal) received from the transceiver circuit 110 into text-only input information for subsequent recognition by using the converted input information. For example, when the input information is voice information, the processor 130 may convert the voice information into the input information only including text by using various speech-to-text (STT) algorithms. In addition, when the input information is image information, the processor 130 may convert the image information into input information including only characters by using a text-to-image algorithm such as an Optical Character Recognition (OCR) algorithm. It should be noted that there is no particular limitation on the above algorithm.

Furthermore, in step S203, the processor 130 may identify one of the context databases, at least one input element and the input intention corresponding to the input information according to the input information, the plurality of context intents and the plurality of sample elements. In other words, the processor 130 may identify the input information, the plurality of contextual intents, and the plurality of sample elements to generate one of the contextual database, the at least one input element, and the input intention corresponding to the input information.

In some embodiments, the memory 120 may store a conversation processing module 1201 and a data processing module 1203, wherein the data processing module 1203 may include a database set 1203 (1), and the database set 1203 (1) may include a plurality of context databases. Further, the dialog processing module 1201 may include an intention recognition module 1201 (1) and a data retrieval module 1201 (3).

Accordingly, the processor 130 may execute the intention recognition module 1201 (1) in the dialog processing module 1201 to perform machine learning (machine learning) calculation on the plurality of contextual intents and the plurality of sample elements to train a recognition model. Accordingly, the processor 130 can use the trained recognition model to perform recognition of one of the context databases, the at least one input element and the input intention in the database 1203 (1) corresponding to the input information.

The processor 130 may execute the intention recognition module 1201 (1) to pre-use a plurality of contextual sentences in a plurality of contextual databases as training samples, and use contextual intents and sample elements corresponding to the contextual sentences as training samples to perform machine learning calculations using the training samples, so as to train a recognition model, wherein the machine learning may be any machine learning calculation for recognition and classification, and there is no particular limitation (for example, dictionary matching calculation or named entity recognition (named entity) calculation may be combined with classification calculation to generate a recognition model for recognizing elements and intents of words and sentences).

Taking the above medical-related databases as an example, the processor 130 may execute the intention recognition module 1201 (1) to generate a recognition model, wherein the recognition model may include a sub-classification model and four sub-recognition modules, and the four sub-recognition modules correspond to the medical information database, the travel warning database, the rumor recognition database, and the epidemic situation distribution prediction database, respectively.

In detail, the processor 130 may execute the intention identifying module 1201 (1) to use the medical query, the travel query, the rumor query, and the epidemic query as the variables, and use the sample elements related to the medical query, the sample elements related to the travel query, the sample elements related to the rumor query, and the sample elements related to the epidemic query as the arguments, so as to perform an arbitrary classification calculation to train a sub-classification model.

Next, the processor 130 may execute the intention recognition module 1201 (1) to use a plurality of word and sentence elements (i.e., a word or a sentence composed of characters) stored in the medical information database as a dependent number and a plurality of sample elements related to the medical query as an argument, and further perform an arbitrary word and sentence recognition operation (e.g., a dictionary matching operation or a named entity recognition operation) to train the first sub-recognition model. In this way, the processor 130 may train a second sub-recognition model, a third sub-recognition model and a fourth sub-recognition model corresponding to the travel warning database, the rumor recognition database and the epidemic situation distribution prediction database, respectively, in the same manner as described above. In this way, the processor 130 may perform the recognition of the input intention by using the sub-classification model, and perform the recognition of the input element by using the four sub-recognition modules.

Therefore, the processor 130 can identify at least one input element corresponding to the input information and an input intention by using the recognition model according to the input information, wherein the input intention can be one of the plurality of contextual intentions. In this way, the processor 130 may execute the data retrieval module 1201 (3) to select one of the context databases in the database set 1203 (1) corresponding to the input intention. The machine learning calculation can also be performed by training a recognition model only by using a plurality of situational intentions to recognize at least one input element and an input intention.

In further embodiments, the dialog processing module 1201 may also include a dialog generation module 1201 (4). When the processor 130 cannot execute the intent recognition module 1201 (1) to recognize at least one input element or input intent corresponding to the input information, the processor 130 may execute the dialog generation module 1201 (4) to output a prompt message (e.g., "please say again") via the transceiver circuit 110 to request the user to provide other input information.

In further embodiments, the data processing module 1203 may also include a data collection module 1203 (2). The processor 130 may execute the data collection module 1203 (2) to periodically update the plurality of sample elements in the plurality of context databases from the external network 300 through the transceiver circuit 110.

In a further embodiment, the processor 130 may execute the user report receiving module 1205 to receive the real-time report message generated from the user device 200 through the transceiver circuit 110 and determine whether to update the context databases according to the real-time report message.

In a further embodiment, the context databases may store a plurality of report forms, and the processor 130 may execute the user report receiving module 1205 to determine whether the real-time report message matches the data format of the report forms. When the processor 130 determines that the real-time report information matches with the data format in any report table, the processor 130 may execute the data collecting module 1203 (2) to update the context databases according to the real-time report information.

In step S205, the processor 130 may determine whether the number of the identified at least one input element is greater than a threshold corresponding to the input intention. When the number of the identified at least one input element is not greater than the threshold value corresponding to the input intention, the process proceeds to step S207. Otherwise, the process proceeds to step S209. In other words, the processor 130 may determine whether the at least one sentence element corresponding to the at least one input element in the corresponding one of the context databases constitutes a feedback information according to the threshold corresponding to the input intention and the at least one input element. When the processor 130 determines that a feedback message cannot be composed, the processor 130 may execute step S207, otherwise, execute step S209.

In some embodiments, the database set 1203 (1) in the memory 120 may pre-store a plurality of thresholds corresponding to a plurality of input intents, respectively. Thereby, the processor 130 may select a threshold value corresponding to the input intention from these threshold values. In a further embodiment, the dialog processing module 1201 may also include a dialog storage module 1201 (2). The processor 130 may store the identified at least one input element and the input intent to the dialog storage module 1201 (2). Accordingly, the processor 130 may execute the dialog storage module 1201 (2) to search the database 1203 (1) for a threshold corresponding to the input intent, thereby determining whether the number of at least one input element is greater than the threshold.

In step S207, the processor 130 may generate at least one prompt element according to the at least one input element and one of the corresponding context databases to generate a prompt message composed of the at least one prompt element, and output the prompt message through the transceiver circuit 110 to request the user to provide other input messages according to the prompt message. Thereby, the step S201 may be returned to. In other words, the processor 130 may generate at least one prompt element related to at least one input element from one of the corresponding context databases to generate a prompt message composed of the at least one prompt element, and output the prompt message through the transceiver circuit 110 to request the user to provide other input messages according to the prompt message. Therefore, the processor 130 can execute step S201 again according to other input information.

In some embodiments, the processor 130 may execute the dialog generating module 1201 (4) to identify the missing element according to the input element and the corresponding sentence element in the context database, and generate a prompt message including a plurality of prompt elements according to the input element and the missing element to request the user to provide other input information according to the prompt message. In a further embodiment, the plurality of context databases in the database set 1203 (1) respectively store a plurality of sentence templates (templates) corresponding to each other. The processor 130 may execute the dialog generation module 1201 (4) to generate a prompt message according to the at least one input element, the at least one missing element, and the statement template corresponding to the at least one input element.

For example, taking the above-mentioned multiple databases related to medical treatment as an example, when the user inputs "physical discomfort" information, the processor 130 may execute the intention recognition module 1201 (1) to recognize that the input intention is a medical query and the input element is "physical" and "discomfort" (e.g., perform any classification operation to recognize the input intention, and perform a dictionary matching operation or a named entity recognition operation to recognize the input element). Next, when the processor 130 executes the dialog generating module 1201 (4) to determine that the input element is smaller than the threshold corresponding to the medical query, the processor 130 may execute the information retrieving module 1201 (3) in the dialog processing module 1201 to search for elements related to the two input elements in the medical information database, thereby generating missing elements of "body parts" and "uncomfortable feeling" (since the medical information database may store various body parts and various feelings related to "uncomfortable", the missing elements may be generated accordingly).

The processor 130 executes the dialog generation module 1201 (4) to generate prompt information including "which body part is uncomfortable and what is uncomfortable" of the plurality of prompt elements according to the input element of "body", "input element of discomfort", "missing element of body part", "missing element of uncomfortable feeling", and corresponding sentence template (corresponding to the input element of "body" and the input element of "discomfort") (for example, according to the input elements and the missing elements, an arbitrary calculation based on similarity-based learning is performed by using at least one sentence template stored in advance in the medical information database, and the prompt information is generated. Therefore, the processor 130 can output the prompt message through the transceiver circuit 110 to request the user to provide other input information according to the prompt message.

In step S209, the processor 130 may search the corresponding one of the context databases for at least one sentence element and at least one sentence template matching the at least one input element. In other words, the processor 130 may search the corresponding one of the context databases for at least one sentence element and at least one sentence template matching the at least one input element. In some embodiments, the processor 130 may execute the information retrieval module 1201 (3) to search the at least one sentence element and the at least one sentence template matching the at least one input element from the corresponding one of the context databases by using any similarity learning based algorithm, which is not particularly limited. In other embodiments, when the processor 130 cannot search for at least one sentence element and at least one sentence template, the processor 130 may execute the information retrieving module 1201 (3) to search for at least one sentence element and at least one sentence template from the external network 300 through the transceiver circuit 110.

Finally, in step S211, the processor 130 may generate feedback information according to the at least one sentence element and the at least one sentence template, so as to output the feedback information through the transceiver circuit 110. In detail, the processor 130 may fill at least one sentence element in the sentence template in a corresponding arrangement to generate the feedback information. In some embodiments, the processor 130 may execute the dialog generation module 1201 (4) to fill at least one sentence element into a corresponding plurality of regions in at least one sentence template, thereby generating the feedback information.

Taking the above-mentioned multiple databases related to medical treatment as an example, when the input intention is a medical query, the processor 130 may execute the information retrieval module 1201 (3) to perform cosine similarity (cosine) calculation, edit distance (edge distance) calculation, manhattan distance (Manhattan distance) calculation, simHash calculation, or the like according to the at least one input element and the medical information database, so as to search the medical information database for the corresponding at least one sentence element and the at least one sentence template. If the input information is "nearby nearest hospital", it can be recognized that the corresponding input intention is a medical query and the input elements are "nearby", "nearest", and "hospital". Based on this, the sentence elements of ChangG Hospital and the sentence template of '8230nearest Hospital' can be searched from the medical information database. Therefore, the feedback information of 'the nearest hospital is ChangG Hospital' can be generated according to the sentence elements and the sentence template.

When the input intent is a travel query, the processor 130 may execute the information retrieval module 1201 (3) to perform Natural Language Processing (NLP) calculation (i.e., semantic recognition) according to the at least one input element and the travel alert database, and then search the travel alert database for at least one corresponding sentence element and at least one sentence template. If the input information is "XX national travel warning in this month", it can be recognized that the corresponding input intention is travel query and the input elements are "this month", "XX country" and "travel warning". Based on the above, the word and sentence elements of 'earthquake' and the sentence template of 'risk of existence of 8230in XX country in this month' can be searched out from the travel warning database, and the feedback information of 'risk of existence of earthquake in XX country in this month' is generated according to the word and sentence elements and the sentence template.

When the intent is a rumor query, the processor 130 may search the corresponding at least one sentence element and at least one sentence template from the rumor identification database according to the same algorithm as the medical query. In addition, the processor 130 may further execute the dialog generation module 1201 (4) to generate a tag (tag) of the rumor label, and further embed the tag when generating the feedback information in the subsequent step (which is beneficial for the user to determine whether to reply the real-time report information according to the tag). If the input information is "true or false for the current month XX state influenza epidemic situation, it can be identified that the corresponding input intention is rumor inquiry and the input elements are" current month "," XX state "and" influenza epidemic situation is out of control ". Based on this, the rumor identification database can be searched out the word and sentence elements of 'influenza epidemic situation is not out of control' and 'the message belongs to the rumor', and the sentence template of 'the XX countries of this month, 8230, and'. Therefore, the feedback information that the current month XX state flu epidemic situation is not out of control and the message belongs to a rumor can be generated according to the sentence element and the sentence template, a label of the rumor is generated, and the label is embedded into the feedback information.

When the input intent is an epidemic query, the processor 130 may also search the corresponding at least one sentence element and at least one sentence template from the epidemic distribution prediction database according to the same algorithm as the medical query. If the input information is "next month XX country influenza epidemic situation prediction", it can be identified that the corresponding input intention is epidemic situation query and the input elements are "next month", "XX country" and "influenza". Based on the above, the sentence elements of ' new york influenza case quantity increase ' and ' next month \8230canbe searched from the epidemic situation distribution prediction database, and the feedback information of ' next month new york influenza case quantity increase and need attention ' is generated according to the sentence elements and the sentence template.

In another example, taking the above-mentioned databases related to enterprise internal information retrieval as an example, when data in the form of tables are stored in the factory information database and the attendance query database, the processor 130 may execute the information retrieval module 1201 (3) to perform table look-up table comparison according to at least one input element and one corresponding context database, and further search one corresponding context database for at least one word element and at least one sentence template.

If the input information is "extension number of 9C01 conference room in a certain office", it can be recognized that the corresponding input intention is the house service inquiry and the input elements are "certain office", "9C01 conference room", and "extension number". Based on this, the sentence element of "5678" and the sentence template of "8230code" extension number can be searched from the factory information database. Therefore, feedback information of 'extension number 5678' can be generated according to the sentence elements and the sentence template. For example, if the input information is "whether the employee number 1234 is on attendance today", it can be identified that the corresponding input intent is an attendance query and the input elements are "employee number", "1234", "today" and "attendance". Based on this, the attendance query database can be searched for "normal" and "8:30 ' and ' attendance status is \8230 ', and attendance time is \8230 ', generating ' attendance status is normal and attendance time is 8:30 ".

In another example, taking multiple databases associated with a plant as an example, the data in the plant information database is the data structure of the question statement corresponding to the answer statement, and the data structure of the data oscillogram in the tool state database. When the input intention is a factory information query, the processor 130 may execute the information retrieval module 1201 (3) to perform cosine similarity calculation, edit distance calculation, manhattan distance calculation, simHash calculation, or the like according to the at least one input element and the factory information database, so as to search the corresponding at least one sentence element and the corresponding at least one sentence template from the factory information database.

If the input information is "query specification of energy storage system PCS2000", it can be identified that the corresponding input intention is factory information query and the input elements are "energy storage system", "PCS2000" and "specification". Based on this, the sentence element of ' technical document ', ' 8230 '; e.g. attached document ' and the corresponding technical document can be searched out from the factory information database, the feedback information of ' technical document such as attached document ' is generated according to the sentence element and the sentence template, and the corresponding technical document is embedded in the feedback information. If the input intent is a tool health status query, the processor 130 may execute the information retrieval module 1201 (3) to search for a corresponding waveform from the tool state database according to the at least one input element, and further identify at least one corresponding sentence element (e.g., abnormal number of tools) and at least one sentence template from the tool state database according to the corresponding waveform. If the input information is "which wafer fabrication tool vibration measurement data are unstable", it can be identified that the corresponding input intent is tool health status query and the input elements are "which", "wafer fabrication tool", "vibration measurement data", and "unstable". Based on this, the corresponding wave pattern can be searched from the machine state database, and then the sentence elements of '1256' and '1380' and the sentence template of 'unstable machine number 8230;' are identified from the machine state database according to the corresponding wave pattern. Therefore, feedback information of unstable machine numbers 1256 and 1380 can be generated according to the sentence elements and the sentence template.

In some embodiments, when the processor 130 cannot search the corresponding one of the context databases for the at least one sentence element or the at least one sentence template matching the at least one input element, the processor 130 may output a prompt message through the transceiver circuit 110 to request the user to provide other input information.

For example, once the processor 130 cannot search one of the context databases for at least one sentence element or at least one sentence template matching at least one input element, the processor 130 may execute the dialog generation module 1201 (4) to generate a prompt message "please re-input once", and then output the prompt message through the transceiver circuit 110 to request the user to perform the input again.

Through the above steps, the query feedback device 100 of the embodiment of the present invention provides a flexible structure, and can determine the input intention and the input elements according to the characters, images, or voices inputted by the user. Thereby, the query feedback device 100 may search various elements similar to the input element from a database corresponding to the input intention to generate feedback information.

Referring to fig. 1 and fig. 4, first, in step S401, the transceiver circuit 110 may receive input information. Next, in step S403, the processor 130 may identify one of the context database, at least one input element and the input intent corresponding to the input information according to the input information, the plurality of context intents and the plurality of sample elements. Finally, in step S405, the processor 130 may generate feedback information corresponding to at least one input element according to one of the corresponding context databases, so as to output the feedback information through the transceiver circuit 110, wherein the one of the context databases corresponds to the input intention.

The query feedback device and the query feedback method provided by the invention can also receive real-time return information of a user so as to further update the database. Although the present invention has been described with reference to the above embodiments, it should be understood that various changes and modifications can be made therein by those skilled in the art without departing from the spirit and scope of the invention.

Claims (15)

1. A query feedback device, comprising:

a transceiver circuit for receiving an input message;

a memory for storing a plurality of instructions and a plurality of context databases, wherein the context databases respectively correspond to a plurality of different context intents, and wherein the context databases have a plurality of sample elements;

a processor coupled to the transceiver circuit and the memory, and configured to execute the instructions:

according to the input information, the situation intents and the sample elements, identifying one situation database, at least one input element and one input intention corresponding to the input information; and

and generating feedback information corresponding to the at least one input element according to the corresponding one of the context databases to output the feedback information through the transceiver circuit, wherein the one of the context databases corresponds to the input intention.

2. The query feedback device as claimed in claim 1, wherein the context databases comprise:

a first context database storing a plurality of first word and sentence elements and a plurality of first sentence templates corresponding to a first context intention; and

and the second situation database stores a plurality of second word and sentence elements and a plurality of second sentence templates corresponding to a second situation intention.

3. The query feedback device of claim 1, wherein the processor is further configured to:

when the input element or the input intention corresponding to the input information cannot be identified, outputting a prompt message through the transceiving circuit to request a user to provide other input information.

4. The query feedback device of claim 1, wherein the processor is further configured to:

when the number of the identified at least one input element is not larger than a threshold value corresponding to the input intention, generating at least one prompt element according to the at least one input element and one of the corresponding situation databases to generate prompt information consisting of the at least one prompt element; and

the prompt message is output through the transceiver circuit to request a user to provide other input information.

5. The query feedback device of claim 1, wherein the processor is further configured to:

performing machine learning calculation on the situation intents and the sample elements to train a recognition model; and

using the recognition model, recognizing one of the context databases, at least one input element and an input intention corresponding to the input information is performed.

6. The query feedback device as claimed in claim 1, wherein the corresponding one of the context databases also corresponds to the input intent, wherein the processor is further configured to:

searching at least one word and sentence element matched with the at least one input element and at least one sentence template from one corresponding situation database; and

generating the feedback information according to the at least one sentence element and the at least one sentence template.

7. The query feedback device of claim 6, wherein the processor is further configured to:

when at least one sentence element or at least one sentence template matched with the at least one input element cannot be searched from the one situation database, a prompt message is output through the transceiving circuit so as to request a user to provide other input information.

8. The query feedback device of claim 1, wherein the processor is further configured to:

and receiving a real-time report message through the transceiver circuit, and judging whether to update the situation databases according to the real-time report message.

9. The query feedback device of claim 1, wherein the processor is further configured to:

the sample elements in the context databases are periodically updated from an external network through the transceiver circuit.

10. A query feedback method, comprising:

receiving an input message through a transceiver circuit;

performing machine learning calculation on a plurality of situation intents or a plurality of situation intents and a plurality of sample elements through a processor to train a recognition model, wherein the sample elements are stored in a plurality of situation databases in a memory, and the situation databases are respectively corresponding to the situation intents;

receiving, by the processor, the input information, the processor performing identifying the at least one input element corresponding to the input information and the input intent using the identification model;

generating, by the processor, a feedback information corresponding to the at least one input element according to one of context databases, wherein the one of context databases corresponds to the input intent; and

the feedback information is output through the transceiver circuit.

11. The query feedback method of claim 10, further comprising:

when the input element or the input intention corresponding to the input information cannot be identified, outputting a prompt message through the transceiving circuit to request a user to provide other input information.

12. The query feedback method of claim 10, further comprising:

when the number of the at least one identified input element is not greater than a threshold corresponding to the input intention, generating at least one prompt element according to the at least one input element and the one context database; and

and outputting a prompt message corresponding to the at least one prompt element through the transceiver circuit to request a user to provide other input information.

13. The query feedback method of claim 10, wherein the step of generating the feedback information corresponding to the at least one input element according to one of the context databases comprises:

searching at least one sentence element matched with the at least one input element and at least one sentence template from the one situation database; and

generating the feedback information according to the at least one sentence element and the at least one sentence template.

14. The query feedback method of claim 13, further comprising:

when the at least one sentence element or the at least one sentence template matched with the at least one input element cannot be searched from one of the context databases, a prompt message is output through the transceiver circuit to request a user to provide other input information.

15. The query feedback method of claim 10, further comprising:

and receiving a real-time report message through the transceiver circuit, and judging whether to update the situation databases according to the real-time report message.

Images