TWI629606B - Dialog method of dialog system - Google Patents

Abstract

A dialog system dialogue method searches for an input question in a community question and answer service, and obtains a plurality of answer data, and then calculates the confidence of each answer data to output the answer data with the highest confidence, due to the answer data Confidence is calculated based on the relationship between the input question and the answer data, which ensures the credibility of the answer data output by the dialogue system.

Description

Dialogue system dialogue method

The present invention relates to a dialog system, and more particularly to a dialog method for a dialog system.

The dialogue system is a computer system for causing a computer to talk to a user. The general dialogue system stores a plurality of training texts, and the training text includes preset questions and corresponding answers. When inputting a question to the dialogue system, the dialogue system first recognizes the text represented by the user's voice, and then calculates the similarity between the user's question and the predicted text of the training text, and compares the highest similarity. Set the answer to the answer to the user, so that the user can get the desired answer through the dialogue system. However, due to the limited number of presupposition questions that can be stored in the training text, the dialogue system can usually only be restricted to a single field, and the grammar and wording of the spoken words are quite variable. When the training text is limited, often Let the user repeat the question before getting the answer.

The main purpose of the present invention is to obtain a plurality of answer data after searching for a community question answering service, and calculate the confidence of each answer data, and finally reply with the answer data with the highest confidence, because Searching for answers in an open database allows the dialog system of the present invention to be not limited to a single field. In addition, confidence is calculated based on the relationship between the input question and the answer data, which ensures the correctness of the answer to the reply. Allow users to get the answers they need.

The dialog method of the dialog system of the present invention comprises: a dialog module receiving an input question, and searching the community question-answering service to obtain a plurality of answer materials; The determining module determines whether the answer data includes structured data, and if so, the determining module controls the dialog module to output the answer data of the structured data, if otherwise, the determining module controls one The calculation module calculates a confidence level of each of the answer data; the calculation module calculates the confidence of each of the answer data according to one of the answer data, an answer order, a similarity, and a keyword density. The calculation of confidence is: among them, For the input question, For the answer, The confidence level for each of the answers to the information, For the number of consents for the input question, Sort the answer to the answer data, For the similarity between the input question and each of the answer materials, For the keyword of the input question, The keyword density for each of the answer questions, For each of the answer data, the number of keywords that contain the input question, For each of the answers, the number of words, , , and It is a variable variable; the dialog module outputs the answer data with the highest confidence.

The invention obtains the answer data by searching the community question and answer service, and calculates the confidence of each answer data by using the calculation module, because the calculation module directly calculates the input question and the answer data. The confidence level can give the user a reliable and credible reply, and since the community question and answer service is an open database, the response of the dialogue system is not limited to a single field, but can be widely applied.

Please refer to FIG. 1 and FIG. 2, which are flowcharts of a dialog method 10 for a dialog system according to an embodiment of the present invention. The dialogue method 10 of the dialog system is used to enable a dialog system 100 to reply to a user's question. Preferably, the dialog method 10 of the dialog system of the embodiment is a Metasearch method, that is, integrating multiple search engines to search for the best reply answer, please refer to FIG. 2, in this embodiment. The dialogue method 10 of the dialogue system is to integrate a search engine SE and a community question answering service (CQA) to answer the answers, but in other embodiments, more search engines and community questions and answers can be used. The service or the search for the answer is only answered by a single community Q&A service CQA, and the present invention is not limited thereto.

Referring to FIG. 2, a user (not shown) asks a speech capture device 110, and the question is retrieved by the speech capture device 110 as a voice signal 111, and the voice signal 111 passes through a voice recognition module. The group 120 is converted into an input question 121, which is in a text format for subsequent processing. In this embodiment, the voice capture device 110 is a microphone, and the voice recognition module 120 converts the voice signal 111 into the input question 121 by Google Automatic Speech Recognition (Google ASR).

Then, referring to FIG. 1 and FIG. 2, step 11 of the dialogue method 10 of the dialog system is performed. The input question 121 is input to a parsing module 130 for analysis. In this embodiment, the parsing module 130 is The CKIP Chinese profiling system developed by the Academia Sinica analyzes the input questions 121 to distinguish the input question 121 into a plurality of vocabularies, wherein the CKIP Chinese profiling system defines at least one vocabulary in the input question 121. For the input phrase 121, the parsing module 130 uses the vocabulary of the vocabulary as the keyword of the input question 121.

Next, referring to FIG. 1 and FIG. 2, step 12 of the dialog method 10 of the dialog system is performed. The dialog module 140 searches the input question 121 on a search engine SE to obtain a plurality of search results SR. In this embodiment, the search engine SE is a Google search system. Please refer to Figures 3 and 4, which are search results SR of the input question 121 of "Taiwan's length" and "Taiwan patent", respectively, Figure 3, "Taiwan's length" search results SR has a selected summary data FS, the selected summary data FS is structured data (usually structured and reliable data, but relatively, please refer to 4, "Taiwan Patent" search results SR does not have structured summary data, the search results are general articles or links to related websites, these search results SR is more difficult to use as a dialogue reply. Therefore, the search results SR are transmitted to the determination module 150 to perform step 13, and the determination module 150 determines whether the search results SR contain structured data. Please refer to the first and third figures, if the search results are obtained. The SR includes the structured data, and the determining module 150 controls the dialog module 140 to directly output the search result SR of the structured data as a reply. Preferably, the search result SR of the structured data is A speech synthesis module (not shown) is converted into speech and played through a playback device (not shown), allowing the user to directly listen to the response of the dialog system 100. Referring to FIGS. 1 and 4, if the search results SR do not include structured data, proceeding to step 15, the dialog module 140 searches the input question 121 for the community question and answer service CQA.

Referring to FIG. 1 and FIG. 2, in step 15, the dialog module 140 searches the input question 121 in a community question answering service CQA to obtain a plurality of answer data AD. In this embodiment, the community question and answer The service CQA is known to Baidu. Please refer to Figures 5 and 6 for the answer data of the input question 121 of “Area in Beijing” and “Alishan Sunrise Time”. Please refer to Figure 5, “Beijing The answer area AD has a recommended answer RA, which is structured data, usually practical and credible, but relatively, please refer to Figure 6, "Alishan Sunrise Time" answer The data AD does not have a structured recommendation answer, and it is not possible to determine which answer data AD is a reliable answer. Therefore, the answer data AD is transmitted to the determination module 150 to perform step 16, and the determination module 150 determines the If the answer data AD includes structured data, please refer to the first and fifth figures. If the answer data AD includes structured data, proceed to step 17, the decision module 150 controls the dialog module 140 to directly output. Structure The answer data AD of the data is used as a reply. Preferably, the answer data AD of the structured data is converted into a voice through the voice synthesis module (not shown), and is played through the playback device (not shown). , allowing the user to directly listen to the reply of the dialogue system 100. Referring to FIGS. 1 and 6, if the answer data AD does not include structured data, step 18 is performed. The determining module 150 controls a computing module 160 to calculate a confidence level of each of the answer data AD.

Referring to FIGS. 1 and 2, the calculation module 160 calculates the confidence of each of the answer data AD according to one of the answer data AD, an answer order, a similarity, and a keyword density. The calculation formula is: among them, For the input question 121, For the answer information AD, For this confidence in the answer data AD, For the consent number of the answer data AD, Sort the answer to the answer data AD, For the similarity between the input question 121 and each of the answer data AD, in the embodiment, the calculation module 160 performs each answer by using a Whole-matching-plane-based (WMPB) method. The calculation of the similarity of the data AD, For the keyword of the input question 121, Ask for the keyword density of AD for each of these answers, For each of the answer materials AD, the number of keywords including the input question 121, Ask each question the number of words in AD, , , and As a variable variable, in this embodiment, , , and They are 0.2, 0.2, 0.2 and 0.4 respectively to achieve the best response accuracy.

Referring to FIG. 1 , after completing the calculation of the confidence level of the answer data AD, the process proceeds to step 19. The dialog module 140 outputs the answer data AD with the highest confidence. Preferably, the answer with the highest confidence is the answer. The data AD is converted into speech through a speech synthesis module (not shown) and played through a playback device (not shown), so that the user can directly listen to the response of the dialogue system 100 to achieve a better person. Machine interaction.

The present invention obtains the answer data by searching the community Q&A service CQA, and calculates the confidence of each of the answer data AD by the calculation module 160, because the calculation module 160 directly uses the input question 121 and Each of the answer data AD calculates its confidence, and can give the user a reliable and credible reply, and since the community Q&A service CQA belongs to the open database, the response of the dialogue system 100 is not limited to a single field, but can be performed. A wide range of applications.

The scope of the present invention is defined by the scope of the appended claims, and any changes and modifications made by those skilled in the art without departing from the spirit and scope of the invention are within the scope of the present invention. .

10‧‧‧ Dialogue system dialogue method
11‧‧‧Analysis of input questions
12‧‧‧Entering a question into a search engine
13‧‧‧Determining whether the search results have structured information
14‧‧‧ Output as search results for structured data
15‧‧‧Searching for questions in the community question and answer service
16‧‧‧Determining whether the answer data has structured information
17‧‧‧ Output as an answer to structured information
18‧‧‧ Calculate the confidence of the answer data
19‧‧‧Output data with the highest confidence
110‧‧‧Voice capture device
111‧‧‧ voice signal
120‧‧‧Voice recognition module
121‧‧‧Input question
130‧‧‧analysis module
140‧‧‧Dialog Module
150‧‧‧Decision module
160‧‧‧Computation Module
CQA‧‧‧Community Q&A Service
SE‧‧ Search Engine
SR‧‧ Search results
FS‧‧‧Selected summary information
AD‧‧‧answer information
RA‧‧‧Recommended answer

Figure 1 is a flow chart of a dialog method for a dialog system in accordance with an embodiment of the present invention. Figure 2: Functional block diagram of a dialog system in accordance with an embodiment of the present invention. Figure 3: Schematic diagram of a search result with structured data for an input question. Figure 4: Schematic diagram of a search result with no structured data for an input question. Figure 5: A schematic diagram of an answer message with structured information entered into it. Figure 6: Schematic diagram of an answer message with no structured information in an input question.

Claims (7)

A dialogue method of a dialogue system, comprising: a dialog module receiving an input question, and searching the community question-answering service to obtain a plurality of answer data; The module determines whether the answer data includes structured data, and if so, the determining module controls the dialog module to output the answer data of the structured data, if otherwise, the determining module controls a calculation The module calculates a confidence level of each of the answer data; the calculation module calculates the confidence level of each of the answer data according to each of the answer data; and the dialog module outputs the answer data with the highest confidence. The dialog method of the dialog system of claim 1, wherein the calculation module calculates each of the answer data according to one of the answer data, one answer order, one similarity, and one keyword density. The confidence level, the calculation formula of the confidence degree is: among them, For the input question, For the answer, The confidence level for each of the answers to the information, For the number of consents for the input question, Sort the answer to the answer data, For the similarity between the input question and each of the answer materials, For the keyword of the input question, The keyword density for each of the answer questions, For each of the answer data, the number of keywords that contain the input question, For each of the answers, the number of words, , , and It is a variable variable. The dialog method of the dialog system of claim 1, wherein the dialog module further comprises a profiling step before the input query, the parsing step is a profiling module parsing the input question, The input question is divided into a plurality of words, and at least one of the words is used as the keyword of the input question. For example, the dialogue method of the dialogue system described in claim 3, wherein the analysis module analyzes the input questions by the CKIP Chinese profiling system developed by the Academia Sinica. The dialog method of the dialog system described in claim 1, wherein the dialog module includes the dialog module to input the query question to a search engine before the input query is searched by the community question and answer service. Performing a search to obtain a plurality of search results, and the determining module determines whether the search results include structured data, and if so, the determining module controls the search result output by the dialog module as structured data, if Otherwise, the dialog module searches the input question in the community question and answer service. For example, in the dialogue method of the dialogue system described in claim 1, wherein the calculation module performs the similarity of each of the answer data by using a whole-matching-plane-based (WMPB) method. Calculation. For example, the dialogue method of the dialogue system described in claim 1 of the patent scope, wherein the confidence calculation formula , , and The size ratio is 1:1:1:2.