WO2017097061A1 - Appareil et procédé de réponse intelligente - Google Patents

Appareil et procédé de réponse intelligente Download PDF

Info

Publication number
WO2017097061A1
WO2017097061A1 PCT/CN2016/104071 CN2016104071W WO2017097061A1 WO 2017097061 A1 WO2017097061 A1 WO 2017097061A1 CN 2016104071 W CN2016104071 W CN 2016104071W WO 2017097061 A1 WO2017097061 A1 WO 2017097061A1
Authority
WO
WIPO (PCT)
Prior art keywords
question
information
type
answer
problem information
Prior art date
Application number
PCT/CN2016/104071
Other languages
English (en)
Chinese (zh)
Inventor
韩丙卫
冯军
丁岩
Original Assignee
中兴通讯股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中兴通讯股份有限公司 filed Critical 中兴通讯股份有限公司
Publication of WO2017097061A1 publication Critical patent/WO2017097061A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/903—Querying
    • G06F16/90335—Query processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation

Definitions

  • the present invention relates to the field of communications, and in particular to an intelligent response method and apparatus.
  • the traditional customer service center is mainly based on the telephone call center.
  • the traditional customer service center can not meet the needs of customers.
  • Many large service companies are constantly expanding their more cost-effective customer service centers, reducing labor costs and enhancing user experience, such as smart online customer service and smartphone applications.
  • the core of intelligent customer service is intelligent question and answer.
  • the question and answer system can get short and accurate answers based on user input.
  • the intelligent question answering system has inflexible semantic support and inconvenient maintenance, and no effective solution has been proposed yet.
  • the embodiment of the invention provides an intelligent response method and device, so as to at least solve the problem that the intelligent question answering system in the related art is inflexible in semantic support and inconvenient to maintain.
  • an intelligent response method includes: receiving problem information input through a user terminal; searching a plurality of question and answer libraries for a question and answer library corresponding to the type of the problem information; The rule searches for an answer corresponding to the question information from a question answering library corresponding to the type of the question information; and returns an answer corresponding to the question information as a return result to the user terminal.
  • searching for the answer corresponding to the question information from the question and answer library corresponding to the type of the question information according to the preset rule includes: selecting the question and answer library corresponding to the type of the question information, according to the question information and The similarity of the question stored in the question and answer library corresponding to the type of the problem information, filtering the question in the question answering library corresponding to the type of the problem information; and the corresponding answer corresponding to the filtered question is corresponding to the problem information s answer.
  • the question information is stored according to the question information and the question and answer library corresponding to the type of the question information.
  • the similarity of the problem, the filtering of the problem in the question answering library corresponding to the type of the problem information includes: obtaining a target problem matching the problem information from the question answering library corresponding to the type of the problem information; obtaining the problem information and the The similarity of the target problem; filtering the target problem in the question and answer library corresponding to the type of the problem information that is less than the preset threshold, and obtaining the first filtering result; when only one target problem is included in the first filtering result, The answer corresponding to the one target question is the answer corresponding to the problem information; when the plurality of the target questions are included in the first filtering result, the target problem corresponding to the highest priority among the plurality of the target questions is taken The answer is the answer to the question information.
  • the method 1 is applied to at least one of the following scenarios: the problem information is the same as the target problem; the problem information includes the target problem; and the target problem includes the problem information.
  • the number of words included in the segmentation of the problem information is obtained by the following rules: segmenting the problem information to obtain a plurality of words; assigning each word according to the part of speech of each word Weighting; the number of each of the plurality of words and the weighted sum of each of the words as the number of words included after the word segmentation of the question; the following rules are used to obtain the word segmentation after the target problem is included
  • the weighted sum of words is the number of words included after the word segmentation of the target question.
  • searching, from a plurality of question and answer libraries, the question and answer library corresponding to the type of the problem information includes: determining whether the problem information is the first type of problem information; and the problem information is the first type of problem In the case of information, obtaining a first type of question and answer library corresponding to the first type of question information; and if the question information is not the first type of question information, determining whether the question information is the second type of question information;
  • the problem information is the second type of problem information
  • the second type of question answering library corresponding to the second type of question information is obtained; from the second type of question answering library, the problem information is reconfirmed according to the preset rule;
  • the reconfirmed question information is the third type of question information; in the case where the reconfirmed question information is the third type of question information, the third type of question answering library corresponding to the reconfirmed question information is obtained.
  • the first type of problem information includes: slang; the first type of question and answer library includes: a slang corpus; wherein the slang corpus stores a slang word and an answer corresponding to the slang word;
  • the second type of problem information includes: specific problem information; the second type of question and answer library includes: an equivalence sentence and a standard sentence library; wherein the equivalence sentence and the standard sentence library store the specific problem and correspond to the specific problem a standard sentence or an equivalent sentence;
  • the third type of problem information includes: general problem information; the third type of question and answer library includes: a general question and an answer corresponding to the ordinary question.
  • reconfirming the problem information according to the preset rule includes: querying, from the second type of question and answer library, all the questions corresponding to the problem information; After the rule is filtered to obtain all the second filtering results, if the second filtering result is unique, the filtering result is used as the re-confirmed problem information; in the case where the second filtering result does not exist Next, the problem information input through the user terminal is used as the problem information after the reconfirmation.
  • the method before returning the answer corresponding to the problem information to the user terminal as a return result, the method further includes: determining whether the answer corresponding to the problem information is the fourth type of problem information; wherein The fourth type of problem information includes: a guided question; if yes, triggering display of the prompt information, wherein the prompt information is used to prompt the user terminal to input the guide information corresponding to the fourth type of problem information; if not, directly The answer corresponding to the question information is returned to the user terminal as a return result.
  • an intelligent response apparatus including: a receiving module configured to receive problem information input through a user terminal; and a first searching module configured to search from a plurality of question and answer libraries a question answering library corresponding to the type of the problem information; the second searching module is configured to search for an answer corresponding to the question information from the question answering library corresponding to the type of the question information according to the preset rule; and returning the module, set to be The answer corresponding to the question information is returned to the user terminal as a return result.
  • the second searching module is further configured to: according to the problem information, the similarity between the problem information and the question stored in the question and answer library corresponding to the type of the problem information, from the question answering library corresponding to the type of the problem information. Filtering the question in the question and answer library corresponding to the type of the problem information; and using the filtered answer corresponding to the question as the answer corresponding to the question information.
  • the second searching module includes: a first acquiring unit, configured to acquire a target problem that matches the problem information from a question and answer library corresponding to the type of the problem information; and the second obtaining unit is configured to Obtaining a similarity between the problem information and the target problem; the filtering unit is configured to filter the target problem in the Q&A library corresponding to the type of the problem information that is less than a preset threshold, to obtain a first filtering result; In order to include only one of the target questions in the first filtering result, the answer corresponding to the one target question is used as an answer corresponding to the problem information; and when the plurality of the target questions are included in the first filtering result, The answer corresponding to the target question with the highest priority among the plurality of the target questions is taken as the answer corresponding to the question information.
  • the first searching module includes: a first determining unit, configured to determine whether the problem information is the first type of problem information; and a third obtaining unit, configured to: the problem information is the first type of problem In the case of information, a first type of question and answer library corresponding to the first type of question information is obtained; and the second determining unit is configured to determine whether the question information is the first if the question information is not the first type of question information.
  • the fourth obtaining unit is configured to obtain a second type of question and answer library corresponding to the second type of question information if the problem information is the second type of question information; and the confirming unit is set to be from the first In the second type of question answering library, the problem information is reconfirmed according to the preset rule; the third determining unit is configured to determine whether the reconfirmed question information is the third type of problem information; and the fifth obtaining unit is set to reconfirm the In the case where the problem information is the third type of problem information, a third type of question answering library corresponding to the reconfirmed problem information is obtained.
  • the confirmation unit includes: a query subunit, configured to query all the questions corresponding to the problem information from the second type of question and answer library; and the filtering subunit is set to follow the preset rule After filtering all the questions, the second filtering result is obtained; the confirmation subunit is set to use the filtering result as the problem information after the reconfirmation if the second filtering result is unique; and in the absence of the second filtering In the case of the result, the problem information input by the user terminal is used as the problem information after the reconfirmation.
  • the device further includes: a determining module, configured to determine whether the answer corresponding to the problem information is the fourth type of problem information; wherein the fourth type of problem information includes: a guided question; a display module, When the answer corresponding to the question information is the fourth type of question information, triggering the display of the prompt information, wherein the prompt information is used to prompt the user terminal to input the guide information corresponding to the fourth type of problem information;
  • the returning module is further configured to directly return the answer corresponding to the question information as a return result to the user terminal if the answer corresponding to the question information is not the fourth type of question information.
  • Embodiments of the present invention also provide a storage medium, which may be configured to store program code for performing the following steps: receiving problem information input through a user terminal; searching for the problem from a plurality of question and answer libraries A question answering library corresponding to the type of information; searching for an answer corresponding to the question information from a question and answer library corresponding to the type of the question information according to a preset rule; and returning the answer corresponding to the question information as a return result to the user terminal.
  • the problem information input through the user terminal is received; from the plurality of question and answer libraries, the question and answer library corresponding to the type of the problem information is searched; and the question and answer library corresponding to the type of the problem information is selected according to the preset rule. Find an answer corresponding to the question information; return the answer corresponding to the question information as a return result to the user terminal.
  • the question and answer library is divided into multiple question and answer libraries. According to the type of the question information input, the answers corresponding to the question information are searched from the question and answer library corresponding to the type of the question information in the plurality of question and answer libraries, and the intelligence in the related technology is solved.
  • the question and answer system has inflexible semantic support and inconvenient maintenance, which improves the flexibility of semantic support and facilitates the maintenance and update of the intelligent question answering system.
  • FIG. 1 is a flowchart 1 of an intelligent response method according to an embodiment of the present invention.
  • FIG. 2 is a second flowchart of an intelligent response method according to an embodiment of the present invention.
  • FIG. 3 is a third flowchart of an intelligent response method according to an embodiment of the present invention.
  • FIG. 4 is a flowchart 4 of an intelligent response method according to an embodiment of the present invention.
  • FIG. 5 is a flow chart of a smart question answering method in accordance with a preferred embodiment of the present invention.
  • FIG. 6 is a schematic flow chart of a query question and answer process according to a preferred embodiment of the present invention.
  • FIG. 7 is a structural block diagram 1 of an intelligent response device according to an embodiment of the present invention.
  • FIG. 8 is a structural block diagram 2 of an intelligent response apparatus according to an embodiment of the present invention.
  • FIG. 9 is a structural block diagram 3 of an intelligent response apparatus according to an embodiment of the present invention.
  • FIG. 10 is a structural block diagram 4 of an intelligent response apparatus according to an embodiment of the present invention.
  • FIG. 11 is a structural block diagram 5 of an intelligent response apparatus according to an embodiment of the present invention.
  • FIG. 12 is a block diagram showing the architecture of a smart question answering system in accordance with a preferred embodiment of the present invention.
  • FIG. 1 is a flowchart 1 of an intelligent response method according to an embodiment of the present invention. As shown in FIG. 1 , the process includes the following steps:
  • Step S102 receiving problem information input through the user terminal
  • Step S104 searching for a question and answer library corresponding to the type of the problem information from the plurality of question and answer libraries;
  • Step S106 searching for an answer corresponding to the problem information from a question and answer library corresponding to the type of the problem information according to a preset rule;
  • step S108 the answer corresponding to the question information is returned to the user terminal as a return result.
  • the question and answer library is divided into multiple, different questions.
  • the type of information corresponds to different question and answer libraries, that is, different question and answer libraries store different types of questions and answers.
  • the type of the question information is from the question and answer library corresponding to the type of the question information.
  • the intelligent question answering system is inflexible in semantic support and inconvenient to maintain.
  • the above problem information may include: “Hello”, “How is the weather” and the like, with certain characteristics such as using dialects to describe how to query the specific fee information, etc., “How to check the phone bill”
  • Ordinary problem information such as "how to send a text message”, a guided question such as "query call charge”
  • the types of the above question and answer library may include: a slang corpus, an equivalent sentence and a standard sentence library, and a general question and answer library;
  • the slang corpus stores a slang word and an answer corresponding to the slang word;
  • the equivalent sentence and the standard sentence library store the specific question and a standard sentence or an equivalent sentence corresponding to the specific question; Stores common questions and answers to the common questions.
  • Each type of question information corresponds to a type of question and answer library.
  • the step S104 may include: selecting, from the question and answer library corresponding to the type of the problem information, the similarity of the problem information and the question stored in the question and answer library corresponding to the type of the question information. Filtering the question in the question and answer library corresponding to the type of the problem information; and using the filtered answer corresponding to the question as the answer corresponding to the question information.
  • the method may be: obtaining a target problem that matches the problem information from a question and answer library corresponding to the type of the problem information; obtaining a similarity between the problem information and the target problem; filtering the type of the problem information In the corresponding question and answer library, the target problem with the similarity less than the preset threshold is obtained, and the first filtering result is obtained; when only one target problem is included in the first filtering result, the answer corresponding to the target problem is taken as the problem The answer corresponding to the information; when the plurality of the target questions are included in the first filtering result, the answer corresponding to the target question with the highest priority among the plurality of the target questions is taken as the answer corresponding to the problem information.
  • the above-mentioned target problem may be a problem related to the problem information
  • the related questions are found in the question and answer library corresponding to the type of the problem information, and the similarity between the problem information and the related problems is calculated.
  • the related problems whose similarity is less than a certain threshold are filtered out. If there is only one related problem after filtering, the answer corresponding to the related question is taken as the answer corresponding to the problem information; if it is filtered out, When multiple related questions are concerned, the answer corresponding to the question with the highest priority may be taken as the answer corresponding to the question information according to the priority of the related questions.
  • the priority may be the ranking score of the search engine of the related question, and the priority of the ranked high is high. That is to find the answer corresponding to the problem information through the combination of search and similarity filtering, and solve the problem that the search engine hits too many results in the related art, can not directly return to the user's answer, and overcomes the common similarity. Calculate the problem of weak semantic support.
  • the preset threshold may be preset, may be a fixed value, or may be adjusted in real time according to actual conditions, and the preset threshold may be set as empirical data.
  • the number of words; words2 is the number of words included after the word segmentation of the target question; Samewords is the number of words included in the word segmentation after the word segmentation is performed after segmentation of the target question; Num1 is the number of words in the Samewords that are segmented
  • the number of words included in the word segmentation of the problem information may be obtained by dividing the problem information into a plurality of words; each word of the word is used for each word.
  • the word assigns a weight; the number of each of the plurality of words and the weighted sum of the words are the number of words included after the word segmentation of the question information;
  • the target problem can be obtained by the following rules The number of words included after the word segmentation: segmentation of the target question, obtaining a plurality of words; assigning a weight to each word according to the part of speech of each word; the number of each word of the plurality of words And the weighted sum of each word as the number of words included after the word segmentation of the target question.
  • the weights of different part-of-speech allocations are different.
  • the weight of a noun can be 2
  • the weight of a verb can be 2
  • the weight of an adjective is 1, wherein a weight of 1 means that when calculating the total number of words included, The word is only calculated once. If the weight is 2, it means that when calculating the total number of words included, the word needs to be calculated twice.
  • mode 2 may be employed, but is not limited thereto.
  • FIG. 2 is a flowchart 2 of an intelligent response method according to an embodiment of the present invention.
  • the foregoing step S104 may include:
  • Step S104-1 determining whether the problem information is the first type of problem information
  • Step S104-2 if the problem information is the first type of problem information, acquiring a first type of question answering library corresponding to the first type of problem information;
  • Step S104-3 if the problem information is not the first type of problem information, determining whether the problem information is the second type of problem information;
  • Step S104-4 if the problem information is the second type of problem information, acquiring a second type of question answering library corresponding to the second type of problem information;
  • Step S104-5 from the second type of question and answer library, re-confirm the problem information according to the preset rule;
  • Step S104-6 determining whether the reconfirmed problem information is the third type of problem information
  • Step S104-7 in the case where the reconfirmed problem information is the third type of question information, acquire a third type of question answering library corresponding to the reconfirmed question information.
  • the first type of problem information may include: the above-mentioned slang language; the above-mentioned first type question and answer
  • the library includes: the above-mentioned slang corpus;
  • the second type of problem information may include: specific problem information;
  • the second type of question and answer library may include: an equivalent sentence and a standard sentence library;
  • the third type of problem information may include: general problem information
  • the third type of question and answer library mentioned above may include: a general question and answer library.
  • the above specific problem may be some personalized problem.
  • the corresponding answer is found by searching for the equivalent sentence or standard sentence corresponding to the equivalent sentence and the standard sentence database. .
  • FIG. 3 is a flowchart of a smart response method according to an embodiment of the present invention.
  • the foregoing step S212 may include:
  • Step S212-1 querying, from the second type of question and answer library, all the questions corresponding to the problem information
  • Step S212-2 after filtering all the questions according to the preset rule, obtaining a second filtering result
  • Step S212-3 in the case that the second filtering result is unique, the filtering result is used as the re-confirmed problem information; and in the case where the second filtering result does not exist, the problem information input through the user terminal is to be As the problem information after the reconfirmation.
  • the preset rule may be the similarity filtering mode described above, and the problem of low similarity is filtered out by setting the threshold value by calculating the similarity between the problem information and all the questions corresponding to the problem information. The remaining question may be the issue of reconfirmation.
  • FIG. 4 is a flowchart of a smart response method according to an embodiment of the present invention. As shown in FIG. 4, before the step S108, the method may further include:
  • Step S402 determining whether the answer corresponding to the problem information is the fourth type of problem information; wherein the fourth type of problem information includes: a guided question;
  • Step S404 if yes, triggering display of the prompt information, wherein the prompt information is used to prompt the user terminal to input the guide information corresponding to the fourth type of problem information;
  • Step S406 if not, directly returning the answer corresponding to the question information to the user terminal as a return result.
  • the above-mentioned guided question may be an extension of the ordinary question and answer.
  • the guided question cannot directly obtain the answer, and must obtain some necessary information through interactive guidance to obtain the final result.
  • the method further includes: determining whether the problem information includes a sensitive word, and if the sensitive word is included, performing step S104 to step S108 instead of directly returning, indicating that the sensitive word is included; If the sensitive word is not included, step S104 is performed.
  • sensitive words can be words such as illegal drugs, yellow-related and other illegal laws.
  • the present invention provides a preferred intelligent question answering method.
  • the user input problem is required.
  • Class, in the intelligent question answering system, the user's input can be divided into the following categories:
  • Ordinary question and answer (equivalent to the general problem information in the above embodiment):
  • the ordinary question and answer is that the user enters a question and wants to directly obtain the answer to the question.
  • the difference with a normal search engine is that the search engine returns multiple answers to a question, requiring the user to select the desired result from multiple answers; and the intelligent question answering system directly returns the most similar answer to the user.
  • Guided question and answer is an extension of the general question and answer. A question cannot be directly obtained. It must be interactively guided to obtain some necessary information before the final result can be obtained. If the user enters "query call charge", this question must guide the user to enter the mobile phone number, month and other information to get the final answer.
  • the intelligent question answering system has different processing logic depending on the type of problem.
  • the question answering library in the intelligent question answering system is divided into the following types, and the corresponding data is saved through the following question and answer library:
  • the chilling database (corresponding to the slang corpus in the above embodiment): the corpus of the chilling dialogue such as the slang and the answer corresponding to the slang, the data is saved to the system index library (corresponding to the slang corpus in the above embodiment).
  • the general question and answer library (equivalent to the general question and answer library in the above embodiment): the general problem and the answer corresponding to the common question, and the corpus of the guided question.
  • Equivalent sentence and standard sentence library (equivalent to the equivalent sentence and standard sentence library in the above embodiment): Save the standard sentence corresponding to the personalized question and the personalized question.
  • the general question and answer library stores The standard question and the answer corresponding to the standard question, this corpus is relatively small, and in the formal intelligent question answering system, the user input problem is in various forms, and the equivalent sentence and the standard sentence library also store the user input problem and Correspondence of standard questions.
  • the above preferred intelligent question and answer method may include the following steps:
  • Step 1 the user enters a question
  • Step 2 The sensitive word determination includes the following steps: Step 21: Determine whether the input contains a sensitive word, if the sensitive word is included, perform step 22, and do not include performing step 31; Step 22, including the sensitive word, directly returning, and not continuing, The prompt contains sensitive words;
  • Step 3 Hanyu judgment: Step 31: Determine whether the user input is a slang word, search from the slang corpus, and if the searched result has a value after threshold filtering, the input question is considered to be a chilling conversation; step 32, if input Is a cold dialogue, choose one of the results to return to the user, do not continue to execute; if it is not a cold dialogue, go to step 41;
  • Step 4 Equivalence sentence search and standard question replacement: Step 41: Using the user input question as a query condition, querying in the equivalence sentence and the standard sentence library, and querying all relevant results; Step 42, performing similarity on the query result Calculate, if there is a unique result after filtering the calculation result, replace the standard question of the result with the user's input question, if not Exist, retain the original problem;
  • Step 5 Search the question and answer library: Step 51, after processing in step 4, search the general question and answer library with the processed question, and query all relevant results; Step 52, perform the similarity calculation on the query result, and if there is a unique result after filtering the calculation result For a result, the result is taken as a question and answer record, and the answer field of the record is taken as a pre-return result.
  • step 6 the process is arranged to determine whether the result is a guided problem: in step 61, after the step 5 is processed, the pre-return result is subjected to the guided question determination; and in step 62, if the guided question has subsequent processing, the process is performed.
  • the arrangement process prompts the user to input; step 63, if it is not a guided question, directly returns the pre-return result as an answer to the user.
  • step 1 corresponds to step S102 in the above embodiment
  • steps 2 to 4 correspond to steps S202 to S214 in the above embodiment
  • step 5 corresponds to step S108
  • step 6 corresponds to the above steps. S402 to step S406.
  • the similarity calculation method may be: parameter 1: user input (corresponding to problem information in the above embodiment) parameter 2: search result (corresponding to the target problem in the above embodiment),
  • the above similarity calculation method includes steps S1 to S6:
  • Step S1 If the user input is exactly the same as the search result, the similarity is 1.0; if the parameter 1 contains the parameter 2, or the parameter 2 includes the parameter 1 to perform the step S2; otherwise, the step S3 is performed.
  • step S2 the longer value of the parameter is length1 and the other is length2, and the similarity score is calculated: alpha1+beta1*length2/length1; the calculation result is returned.
  • the alpha1 and beta1 are empirical values.
  • Step S3 performing word segmentation and part-of-speech tagging on user input and search results.
  • Step S4 adjusting the weight after the word segmentation, the noun and the verb are adjusted to 2, and the other parts of speech are unchanged.
  • Step S5 the number of the same words after the statistical parameter 1 and the parameter 2 word segmentation, the weight is 2, calculated twice, and the count is samewords.
  • Step S6 after the parameter 1 word segmentation is words1, after the parameter 2 word segmentation is words2, the samewords contains the number of words in the word1 is num1, and the samewords contains the number of words in the words2 is num2.
  • the algorithm for specific similarity is:
  • Double dp Math.min(1.0*words1.size()/words2.size(),1.0*words2.size()/words1.size());
  • Double part1 alpha*(1.0*samewords.size()/words1.size()+1.0*samewords.size()/words2.size())/2.0;
  • Double part2 beta*dp*(num1/words1.size()+num2/words2.size())/2.0;
  • the similarity is the value of part1+part2, where size() in the preferred embodiment is used to calculate the number of elements.
  • the threshold filtering method includes the following steps 1 to 4:
  • Step 1 calculating the phase of the input question and the search hit result problem (corresponding to the target problem in the above embodiment) field Similarity
  • Step 3 If after step 1, 2, only one problem result remains after processing, the result is the final result, and the answer field value is returned from the result. If the result of the problem after the processing of steps 1 and 2 is greater than one, the remaining results are filtered by the ranking of the search engine searched from the question and answer library, and the item with the highest sorting score (scoreMax) is taken as the final result. Filter out the remaining problem results.
  • Step 4 After the processing in step 3, if the final result is one, the result is the final result, and the answer field value is returned from the result.
  • the smart question answering method includes:
  • Step S502 the user inputs a question
  • Step S504 it is determined whether the session is set (session); if not, step S506 is performed; in the case of YES, step S508 is performed;
  • Step S506 performing a normal search
  • Step S508 the robot searches whether the question input by the user contains a sensitive word; if it is included, step S510 is performed; if not, step S512 is performed;
  • Step S510 returning directly, reminding the user to include the sensitive word
  • Step S512 normalizing the problem input by the user, such as special symbol processing
  • Step S514 it is determined whether the session exists and there is a cache; if yes, step S516 is performed; if not, step S522 is performed;
  • Step S5128 it is determined whether the guided problem is processed normally, in the case of normal processing, step S520 is performed; if the processing is not normal, step S522 is performed;
  • Step S520 caching the problem, and returning the result
  • Step S522 it is determined whether the problem is a slang; if yes, step S524 is performed; if not, step S526 is performed;
  • Step S530 it is determined whether it is a guided question, if yes, step S536 is performed; if not, step S538 is performed;
  • Step S532 determining context information; performing step S538;
  • step S540 the result is returned.
  • step S526 can be implemented by the following process. As shown in FIG. 6, the process includes steps S602 to S626:
  • Step S602 accurately querying the equivalent sentence in the equivalent sentence and the standard sentence library
  • Step S604 it is determined whether an equivalence sentence is found, if yes, step S606 is performed; if not, step S608 is performed;
  • Step S606 performing standard sentence replacement
  • Step S608 using regular expression matching
  • Step S610 it is determined whether the unique result is matched; if yes, step S606 is performed;
  • Step S612 searching for the standard sentence in the general question and answer library
  • Step S614 performing similarity filtering
  • Step S616 it is determined whether there is a problem result corresponding to the standard sentence after the similarity filtering; if not, step S618 is performed; if yes, step S626 is performed;
  • Step S618, the word segmentation query standard sentence; the standard sentence is divided into words, and the problem corresponding to the standard sentence is segmented;
  • Step S620 performing similarity filtering
  • Step S622 it is determined whether there is a unique problem result corresponding to the standard sentence after filtering; if yes, step S624 is performed;
  • Step S624 searching the general question and answer library to query the answer
  • step S626 the number of answers is obtained.
  • the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware, but in many cases, the former is A better implementation.
  • the technical solution of the present invention which is essential or contributes to the prior art, may be embodied in the form of a software product stored in a storage medium (such as ROM/RAM, disk, CD-ROM, including a number of instructions to make a terminal device (can be a mobile phone, a computer, The server, or network device, etc.) performs the methods described in various embodiments of the present invention.
  • module may implement a combination of software and/or hardware of a predetermined function.
  • apparatus described in the following embodiments is preferably implemented in software, hardware, or a combination of software and hardware, is also possible and contemplated.
  • FIG. 7 is a structural block diagram 1 of an intelligent response apparatus according to an embodiment of the present invention. As shown in FIG. 7, the apparatus includes:
  • the receiving module 72 is configured to receive the problem information input through the user terminal;
  • the first searching module 74 is connected to the receiving module 72, and is configured to search for a question and answer library corresponding to the type of the problem information from the plurality of question and answer libraries;
  • the second searching module 76 is connected to the first searching module 74, and is configured to search for an answer corresponding to the problem information from a question and answer library corresponding to the type of the problem information according to a preset rule.
  • the returning module 78 is connected to the second searching module 76, and is configured to return the answer corresponding to the question information to the user terminal as a return result.
  • the first searching module 74 and the second searching module 76 search for the problem information from the question and answer library corresponding to the type of the problem information.
  • the answer corresponding to the problem information that is, by dividing the question and answer library into multiples, the types of different problem information correspond to different question and answer libraries, that is, different question and answer libraries store different types of questions and answers, compared with the intelligence in related technologies.
  • the response system enhances the semantic support ability and improves the speed of question and answer.
  • different question and answer libraries store different types of questions and answers, it is more convenient to solve the problematic library for targeted update and maintenance.
  • the intelligent question answering system is inflexible in semantic support and inconvenient to maintain.
  • the above problem information may include: “Hello”, “How is the weather” and the like, with certain characteristics such as using dialects to describe how to query the specific fee information, etc., “How to check the phone bill”
  • Ordinary problem information such as "how to send a text message”, a guided question such as "query call charge”
  • the types of the above question and answer library may include: a slang corpus, an equivalent sentence and a standard sentence library, and a general question and answer library;
  • the slang corpus stores a slang word and an answer corresponding to the slang word;
  • the equivalent sentence and the standard sentence library store the specific question and a standard sentence or an equivalent sentence corresponding to the specific question; Stores common questions and answers to the common questions.
  • Each type of question information corresponds to a type of question and answer library.
  • the second searching module 76 is further configured to: according to the question information and the question stored in the question and answer library corresponding to the type of the question information, from the question answering library corresponding to the type of the question information.
  • the similarity degree is filtered by the question and answer library corresponding to the type of the problem information; the filtered answer corresponding to the question is used as the answer corresponding to the question information.
  • FIG. 8 is a structural block diagram of an intelligent response apparatus according to an embodiment of the present invention. As shown in FIG. 8, the second searching module 76 includes:
  • the first obtaining unit 82 is configured to obtain, from the question and answer library corresponding to the type of the problem information, the problem information Target problem
  • the second obtaining unit 84 is connected to the first acquiring unit 82, and is configured to acquire the similarity between the problem information and the target problem;
  • the filtering unit 86 is connected to the second obtaining unit 84, and is configured to filter a target problem in the Q&A library corresponding to the type of the problem information that is less than a preset threshold, to obtain a first filtering result;
  • the searching unit 88 is connected to the filtering unit 86, and is configured to: when the first filtering result includes only one target question, the answer corresponding to the one target question is used as an answer corresponding to the problem information; and at the first When the plurality of the target questions are included in the filtering result, the answer corresponding to the target question with the highest priority among the plurality of the target questions is taken as the answer corresponding to the problem information.
  • the foregoing target problem may be a problem related to the problem information
  • the second searching module 76 finds related problems in the question and answer library corresponding to the type of the problem information, and calculates the problem information and the correlation.
  • the similarity of the problem first filter out the related problems whose similarity is less than a certain threshold. If there is only one related problem after filtering, the answer corresponding to the related question is taken as the answer corresponding to the problem information; If there are multiple related problems after filtering out, the answers corresponding to the questions with the highest priority can be taken as the answers corresponding to the problem information according to the priority of these related questions.
  • the priority may be the ranking score of the search engine of the related question, and the priority of the sorted high is high.
  • the preset threshold may be preset, may be a fixed value, or may be adjusted in real time according to actual conditions, and the preset threshold may be set as empirical data.
  • the second obtaining unit 84 may obtain the number of words included after the word segmentation of the problem information by using the following rule: segmenting the problem information to obtain a plurality of words; each of the words according to the part of the word a word assigning weight; the number of each of the plurality of words and the weighted sum of each of the words as the number of words included after the word segmentation of the question information; the word segmentation of the target question can be obtained by the following rules
  • Each The weighted sum of words is the number of words included after the word segmentation of the target question.
  • the weights of different part-of-speech allocations are different.
  • the weight of a noun can be 2
  • the weight of a verb can be 2
  • the weight of an adjective is 1, wherein a weight of 1 means that when calculating the total number of words included, The word is only calculated once. If the weight is 2, it means that when calculating the total number of words included, the word needs to be calculated twice.
  • mode 2 may be employed, but is not limited thereto.
  • FIG. 9 is a structural block diagram 3 of an intelligent response apparatus according to an embodiment of the present invention.
  • the first search module 74 may include:
  • the first determining unit 92 is configured to determine whether the problem information is the first type of problem information
  • the third obtaining unit 94 is connected to the first determining unit 92, and is configured to acquire a first type of question answering library corresponding to the first type of question information if the problem information is the first type of question information;
  • the second determining unit 96 is connected to the third acquiring unit 94, and is configured to determine whether the problem information is the second type of problem information if the problem information is not the first type of problem information;
  • the fourth obtaining unit 98 is connected to the second determining unit 96, and is configured to acquire a second type of question answering library corresponding to the second type of question information if the question information is the second type of question information;
  • the confirmation unit 910 is connected to the fourth obtaining unit 98, and is configured to re-confirm the problem information according to the preset rule from the second type of question and answer library;
  • the third determining unit 912 is connected to the confirming unit 910, and is configured to determine whether the reconfirmed question information is the third type of problem information;
  • the fifth obtaining unit 914 is connected to the third determining unit 912, and is configured to acquire a third type of question answering library corresponding to the reconfirmed question information if the reconfirmed question information is the third type of question information.
  • the first type of problem information may include: the above-mentioned slang; the first type of question and answer library may include: the slang corpus; the second type of problem information may include: specific problem information;
  • the quiz library may include: an equivalence sentence and a standard sentence library;
  • the third type of problem information may include: general problem information;
  • the third type of question and answer library may include: a general question and answer library.
  • the above specific problem may be some personalized problem.
  • the corresponding answer is found by searching for the equivalent sentence or standard sentence corresponding to the equivalent sentence and the standard sentence database. .
  • FIG. 10 is a block diagram showing the structure of an intelligent response device according to an embodiment of the present invention.
  • the confirmation unit 910 may include:
  • the query subunit 1002 is configured to query all questions corresponding to the problem information from the second type of question and answer library;
  • the filtering subunit 1004 is connected to the query subunit 1002, and is configured to filter all the questions according to the preset rule to obtain a second filtering result.
  • the confirmation subunit 1006 is connected to the filtering subunit 1004, and is configured to use the filtering result as the problem information after the reconfirmation when the second filtering result is unique; and in the case where the second filtering result does not exist Next, the problem information input through the user terminal is used as the problem information after the reconfirmation.
  • the preset rule may be the similarity filtering mode described above, and the problem of low similarity is filtered out by setting the threshold value by calculating the similarity between the problem information and all the questions corresponding to the problem information. The remaining question may be the issue of reconfirmation.
  • FIG 11 is a block diagram showing the structure of an intelligent response device according to an embodiment of the present invention. As shown in Figure 11, the device further includes:
  • the determining module 1102 is configured to determine whether the answer corresponding to the problem information is the fourth type of problem information; wherein the fourth type of problem information includes: a guided question;
  • the display module 1104 is connected to the determining module 1102, and is configured to trigger the display of the prompt information when the answer corresponding to the question information is the fourth type of question information, wherein the prompt information is used to prompt the user terminal to input and Guidance information corresponding to the fourth type of problem information;
  • the returning module 78 is further configured to directly return the answer corresponding to the question information as a return result to the user terminal if the answer corresponding to the question information is not the fourth type of question information.
  • the above-mentioned guided question may be an extension of the ordinary question and answer.
  • the guided question cannot directly obtain the answer, and must obtain some necessary information through interactive guidance to obtain the final result.
  • the foregoing apparatus may further include another determining module, and is connected to the receiving module 72, and is configured to determine whether the problem information includes a sensitive word, and if the sensitive word is included, the answer corresponding to the problem information is not queried. Instead, it returns directly, the prompt contains sensitive words; if the sensitive words are not included, the first search module 74, the second search module 76, and the return module 78 continue to work.
  • sensitive words can be words such as illegal drugs, yellow-related and other illegal laws.
  • FIG. 12 is a schematic structural diagram of a smart question answering system according to a preferred embodiment of the present invention.
  • the smart question answering system includes: a management platform 1202, Search engine module 1204, interactive question and answer service module 1206;
  • the interactive question answering service module 1206 includes: a sensitive word filtering module 1208, configured to determine a user input question. Whether the sensitive word is included; the greeting dialog module 1210 (corresponding to the first determining unit 92 in the above embodiment) is configured to determine whether the question input by the user is a slang; the equivalent sentence and the standard sentence conversion module 1212 (equivalent to the above implementation)
  • the second judging unit 96 and the fourth obtaining unit 98) in the example are configured to convert the question input by the user into an equivalent sentence or a standard sentence; the regular expression matching module 1214 is set to not find a problem corresponding to the question input by the user.
  • the regular input matching is performed on the question input by the user;
  • the participle part-of-speech tagging module 1216 is set to segment the question input by the user, and the part of the word after the word segmentation is marked, and/or, the searched As a result, the word segmentation is performed, and the part of speech of the word after the word segmentation is marked;
  • the similarity calculation module 1218 (corresponding to the filter subunit 1004 in the above embodiment) is set to calculate the similarity between the user input question and the searched problem result, wherein
  • the flow programming module 1220 is set to perform process scheduling processing when the problem input by the user is a guided problem, and prompt the user to input relevant guiding information.
  • each of the above modules may be implemented by software or hardware.
  • the foregoing may be implemented by, but not limited to, the foregoing modules are all located in the same processor; or, the modules are located in multiple In the processor.
  • Embodiments of the present invention also provide a storage medium.
  • the foregoing storage medium may be configured to store program code for performing the following steps:
  • the foregoing storage medium may include, but not limited to, a USB flash drive, a Read-Only Memory (ROM), a Random Access Memory (RAM), a mobile hard disk, and a magnetic memory.
  • ROM Read-Only Memory
  • RAM Random Access Memory
  • a mobile hard disk e.g., a hard disk
  • magnetic memory e.g., a hard disk
  • modules or steps of the present invention described above can be implemented by a general-purpose computing device that can be centralized on a single computing device or distributed across a network of multiple computing devices. Alternatively, they may be implemented by program code executable by the computing device such that they may be stored in the storage device by the computing device and, in some cases, may be different from the order herein.
  • the steps shown or described are performed, or they are separately fabricated into individual integrated circuit modules, or a plurality of modules or steps thereof are fabricated as a single integrated circuit module.
  • the invention is not limited to any specific combination of hardware and software.
  • the problem information input through the user terminal is received; from the plurality of question and answer libraries, the question and answer library corresponding to the type of the problem information is searched; and the information related to the problem is obtained according to the preset rule.
  • the answer corresponding to the question information is searched in the question answering library corresponding to the type; the answer corresponding to the question information is returned as a return result to the user terminal.
  • the question and answer library is divided into multiple question and answer libraries. According to the type of the question information input, the answers corresponding to the question information are searched from the question and answer library corresponding to the type of the question information in the plurality of question and answer libraries, and the intelligence in the related technology is solved.
  • the question and answer system has inflexible semantic support and inconvenient maintenance, which improves the flexibility of semantic support and facilitates the maintenance and update of the intelligent question answering system.

Abstract

L'invention concerne un procédé et un appareil de réponse intelligente. Le procédé consiste à : recevoir des informations de question entrées par un terminal utilisateur (S102) ; consulter une pluralité de bibliothèques de questions et de réponses pour trouver une bibliothèque de questions et de réponses correspondant au type des informations de question (S104) ; rechercher, dans la bibliothèque de questions et de réponses correspondant au type des informations de question, une réponse correspondant aux informations de question selon une règle prédéfinie (S106) ; et renvoyer une réponse correspondant aux informations de question au terminal utilisateur en tant que résultat de retour (S108). Le procédé résout les problèmes, rencontrés dans l'état de la technique, liés au support sémantique non flexible d'un système de questions-réponses intelligent et à sa maintenance incommode, ce qui améliore la flexibilité du support sémantique, et maintient et met à jour de manière pratique le système de questions-réponses intelligent.
PCT/CN2016/104071 2015-12-07 2016-10-31 Appareil et procédé de réponse intelligente WO2017097061A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510891197.7A CN106844400A (zh) 2015-12-07 2015-12-07 智能应答方法及装置
CN201510891197.7 2015-12-07

Publications (1)

Publication Number Publication Date
WO2017097061A1 true WO2017097061A1 (fr) 2017-06-15

Family

ID=59013710

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/104071 WO2017097061A1 (fr) 2015-12-07 2016-10-31 Appareil et procédé de réponse intelligente

Country Status (2)

Country Link
CN (1) CN106844400A (fr)
WO (1) WO2017097061A1 (fr)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108509609A (zh) * 2018-04-03 2018-09-07 广州幽联信息技术有限公司 智能人机交互方法、装置、计算机设备及存储介质
WO2019011356A1 (fr) * 2017-07-14 2019-01-17 Cognigy Gmbh Procédé de conduite de dialogue homme-ordinateur
CN113409907A (zh) * 2021-07-19 2021-09-17 广州方舟信息科技有限公司 一种基于互联网医院的智能预问诊方法及系统

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105893552B (zh) * 2016-03-31 2020-05-05 成都晓多科技有限公司 数据处理方法及装置
CN108446295B (zh) * 2018-01-23 2021-02-12 深圳市阿西莫夫科技有限公司 信息检索方法、装置、计算机设备和存储介质
CN108573046B (zh) * 2018-04-18 2021-06-29 什伯（上海）智能技术有限公司 一种基于ai系统的用户指令处理方法及装置
CN108664644A (zh) * 2018-05-16 2018-10-16 微梦创科网络科技（中国）有限公司 一种问答系统构建方法、问答处理方法及装置
CN109002434A (zh) * 2018-05-31 2018-12-14 青岛理工大学 客服问答匹配方法、服务器及存储介质
CN111382247B (zh) * 2018-12-29 2023-07-14 深圳市优必选科技有限公司 一种内容推送优化方法、内容推送优化装置及电子设备
CN109829048B (zh) * 2019-01-23 2023-06-23 平安科技（深圳）有限公司 电子装置、访谈辅助方法和计算机可读存储介质
CN110110133B (zh) * 2019-04-18 2020-08-11 贝壳找房(北京)科技有限公司 一种智能语音数据生成方法及装置
CN110059172B (zh) * 2019-04-19 2021-09-21 北京百度网讯科技有限公司 基于自然语言理解的推荐答案的方法和装置
CN111831800A (zh) * 2019-08-13 2020-10-27 北京嘀嘀无限科技发展有限公司 问答交互方法、装置、设备及存储介质
CN113704434A (zh) * 2021-09-01 2021-11-26 内蒙古大学 知识库问答方法、电子设备及可读存储介质
CN117473069B (zh) * 2023-12-26 2024-04-12 深圳市明源云客电子商务有限公司 业务语料生成方法、装置、设备及计算机可读存储介质

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6766320B1 (en) * 2000-08-24 2004-07-20 Microsoft Corporation Search engine with natural language-based robust parsing for user query and relevance feedback learning
CN101118554A (zh) * 2007-09-14 2008-02-06 中兴通讯股份有限公司 智能交互式问答系统及其处理方法
CN103019407A (zh) * 2012-11-22 2013-04-03 百度国际科技（深圳）有限公司 输入法应用方法、自动问答处理方法及电子设备、服务器
CN103761334A (zh) * 2014-02-17 2014-04-30 网之易信息技术（北京）有限公司 从题库中查找匹配问题的方法和设备
CN104573000A (zh) * 2015-01-07 2015-04-29 北京云知声信息技术有限公司 基于排序学习的自动问答装置及方法
CN104598445A (zh) * 2013-11-01 2015-05-06 腾讯科技（深圳）有限公司 自动问答系统和方法
CN104657346A (zh) * 2015-01-15 2015-05-27 深圳市前海安测信息技术有限公司 智能交互系统中的问题匹配方法和系统

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100416570C (zh) * 2006-09-22 2008-09-03 浙江大学 一种基于问答库的中文自然语言问答方法
CN103425640A (zh) * 2012-05-14 2013-12-04 华为技术有限公司 一种多媒体问答系统及方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6766320B1 (en) * 2000-08-24 2004-07-20 Microsoft Corporation Search engine with natural language-based robust parsing for user query and relevance feedback learning
CN101118554A (zh) * 2007-09-14 2008-02-06 中兴通讯股份有限公司 智能交互式问答系统及其处理方法
CN103019407A (zh) * 2012-11-22 2013-04-03 百度国际科技（深圳）有限公司 输入法应用方法、自动问答处理方法及电子设备、服务器
CN104598445A (zh) * 2013-11-01 2015-05-06 腾讯科技（深圳）有限公司 自动问答系统和方法
CN103761334A (zh) * 2014-02-17 2014-04-30 网之易信息技术（北京）有限公司 从题库中查找匹配问题的方法和设备
CN104573000A (zh) * 2015-01-07 2015-04-29 北京云知声信息技术有限公司 基于排序学习的自动问答装置及方法
CN104657346A (zh) * 2015-01-15 2015-05-27 深圳市前海安测信息技术有限公司 智能交互系统中的问题匹配方法和系统

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019011356A1 (fr) * 2017-07-14 2019-01-17 Cognigy Gmbh Procédé de conduite de dialogue homme-ordinateur
US11315560B2 (en) 2017-07-14 2022-04-26 Cognigy Gmbh Method for conducting dialog between human and computer
CN108509609A (zh) * 2018-04-03 2018-09-07 广州幽联信息技术有限公司 智能人机交互方法、装置、计算机设备及存储介质
CN113409907A (zh) * 2021-07-19 2021-09-17 广州方舟信息科技有限公司 一种基于互联网医院的智能预问诊方法及系统

Also Published As

Publication number Publication date
CN106844400A (zh) 2017-06-13

Similar Documents

Publication Publication Date Title
WO2017097061A1 (fr) Appareil et procédé de réponse intelligente
US11397772B2 (en) Information search method, apparatus, and system
CN107609101B (zh) 智能交互方法、设备及存储介质
US11669698B2 (en) Method and system for automatic formality classification
CN107818781B (zh) 智能交互方法、设备及存储介质
US10270791B1 (en) Search entity transition matrix and applications of the transition matrix
CN107039040B (zh) 语音识别系统
US20210319051A1 (en) Conversation oriented machine-user interaction
CN107797984B (zh) 智能交互方法、设备及存储介质
EP3617952A1 (fr) Procédé, appareil et système de recherche d'informations
US9881010B1 (en) Suggestions based on document topics
CN110888990B (zh) 文本推荐方法、装置、设备及介质
US10346546B2 (en) Method and system for automatic formality transformation
CN104933081A (zh) 一种搜索建议提供方法及装置
CN109299245B (zh) 知识点召回的方法和装置
CN110415679B (zh) 语音纠错方法、装置、设备和存储介质
CN109597874B (zh) 信息推荐方法、装置及服务器
CN110147494B (zh) 信息搜索方法、装置，存储介质及电子设备
KR101541306B1 (ko) 컴퓨터 실행 가능한 중요 키워드 추출 방법, 이를 수행하는 중요 키워드 추출 서버 및 이를 저장하는 기록매체
CN112926308B (zh) 匹配正文的方法、装置、设备、存储介质以及程序产品
JP2020512651A (ja) 検索方法、装置及び非一時的コンピュータ読取可能記憶媒体
CN113204953A (zh) 基于语义识别的文本匹配方法、设备及设备可读存储介质
CN107665442B (zh) 获取目标用户的方法及装置
CN109977292A (zh) 搜索方法、装置、计算设备和计算机可读存储介质
CN115062135A (zh) 一种专利筛选方法与电子设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16872253

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16872253

Country of ref document: EP

Kind code of ref document: A1