CN107015983A

CN107015983A - A kind of method and apparatus for being used in intelligent answer provide knowledge information

Info

Publication number: CN107015983A
Application number: CN201610056357.0A
Authority: CN
Inventors: 吴庚水
Original assignee: Alibaba Group Holding Ltd
Current assignee: Advanced New Technologies Co Ltd; Advantageous New Technologies Co Ltd
Priority date: 2016-01-27
Filing date: 2016-01-27
Publication date: 2017-08-04

Abstract

The purpose of the application is to provide a kind of method and apparatus for being used in intelligent answer provide knowledge information.Specifically, the identification object reference problem corresponding with the primal problem that user submits in reference to problem base；Corresponding knowledge information is determined according to the object reference problem and question and answer mapping model；The knowledge information is provided to the user.Compared with prior art, the application identification object reference problem corresponding with the primal problem that user submits in reference to problem base, the corresponding knowledge information of object reference problem is determined according to question and answer mapping model, and provide the knowledge information to the user, so that intelligent answer independent of search engine or correlation technique can accurately obtain required for solution or knowledge point, lift the solution coverage rate and accuracy of intelligent answer, further, this programme is by carrying out the accuracy that Semantic Discrimination and completion also further increasing scheme or knowledge Point matching to enquirement side in intelligent answer.

Description

A kind of method and apparatus for being used in intelligent answer provide knowledge information

Technical field

The application is related to computer realm, more particularly to a kind of technology for being used to provide knowledge information.

Background technology

With the development of big data, intelligent use is progressively popularized, and user is helped for convenience, real Existing problem is quickly positioned or faster obtains corresponding solution, and existing software and platform service are progressively adopted Use intelligent Answer System.However, request content of the existing intelligent Answer System according to user, is utilized The intelligent searching engine of customer service system searches for data in knowledge base, and according to confidence level or letter The heart point output result excessively depends on the accuracy of search engine to user, the solution of problem Coverage rate is not high, and the accuracy of question and answer is also limited.

The content of the invention

The purpose of the application is to provide a kind of method and apparatus for being used to provide knowledge information, to solve Knowledge information provides the problem of accuracy rate is limited caused by intelligent Answer System relies on search engine.

To achieve the above object, this application provides a kind of method for providing knowledge information, the party Method solves knowledge information caused by intelligent Answer System relies on search engine and provides what accuracy rate was limited Problem, this method includes：

The identification object reference problem corresponding with the primal problem that user submits in reference to problem base；

Corresponding knowledge information is determined according to the object reference problem and question and answer mapping model；

The knowledge information is provided to the user.

To achieve the above object, this application provides a kind of equipment for providing knowledge information, this sets There is provided what accuracy rate was limited for knowledge information caused by intelligent Answer System relies on search engine is solved Problem, the equipment includes：

Object reference problem identification device, it is original with user's submission for being recognized in reference to problem base The corresponding object reference problem of problem；

Question and answer mapping device, for determining correspondence according to the object reference problem and question and answer mapping model Knowledge information；

Knowledge provides managing device, for the knowledge information to be provided to the user.

Compared with prior art, the application primal problem that identification is submitted with user in reference to problem base Corresponding object reference problem, the corresponding knowledge of object reference problem is determined according to question and answer mapping model Information, and the knowledge information is provided to the user, so that intelligent answer is independent of search Engine or correlation technique can accurately obtain required for solution or knowledge point, lifting intelligence asks The solution coverage rate and accuracy answered, further, this programme pass through to being putd question in intelligent answer Fang Jinhang Semantic Discriminations and completion also further increasing the accuracy of scheme or knowledge Point matching.

Brief description of the drawings

By reading the detailed description made to non-limiting example made with reference to the following drawings, this Shen Other features, objects and advantages please will become more apparent upon：

Fig. 1 shows a kind of method flow for being used to provide knowledge information according to the application one side Figure；

Fig. 2 shows the method flow diagram of the step S1 according to one preferred embodiment of the application；

Fig. 3 shows a kind of method flow for being used to provide knowledge information according to the application other side Figure；

Fig. 4 shows a kind of equipment signal for being used to provide knowledge information according to the application other side Figure；

Fig. 5 shows setting according to the object reference problem identification device of the application another preferred embodiment Standby schematic diagram；

Fig. 6 shows a kind of equipment signal for being used to provide knowledge information according to the application other side Figure；

Fig. 7 shows the schematic diagram of the offer knowledge information according to the application another preferred embodiment.

Same or analogous reference represents same or analogous part in accompanying drawing.

Embodiment

The application is described in further detail below in conjunction with the accompanying drawings.

In one typical configuration of the application, terminal, the equipment of service network and trusted party include One or more processors (CPU), input/output interface, network interface and internal memory.

Internal memory potentially includes the volatile memory in computer-readable medium, random access memory And/or the form, such as read-only storage (ROM) or flash memory (flash such as Nonvolatile memory (RAM) RAM).Internal memory is the example of computer-readable medium.

Computer-readable medium includes permanent and non-permanent, removable and non-removable media can be with Realize that information is stored by any method or technique.Information can be computer-readable instruction, data knot Structure, the module of program or other data.The example of the storage medium of computer includes, but are not limited to phase Become internal memory (PRAM), static RAM (SRAM), dynamic random access memory (DRAM), other kinds of random access memory (RAM), read-only storage (ROM), electricity It is Erasable Programmable Read Only Memory EPROM (EEPROM), fast flash memory bank or other memory techniques, read-only Compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storages, Magnetic cassette tape, magnetic disk storage or other magnetic storage apparatus or any other non-transmission medium, The information that can be accessed by a computing device available for storage.Defined according to herein, it is computer-readable Medium not include non-temporary computer readable media (transitory media), such as modulation data-signal and Carrier wave.

Fig. 1 shows a kind of method flow for being used to provide knowledge information according to the application one side Figure.Methods described includes step S1, step S2, step S3.

Wherein, the equipment 1 recognizes the original submitted with user in reference to problem base in step sl The corresponding object reference problem of beginning problem；The equipment 1 is joined according to the target in step s 2 Examination topic and question and answer mapping model determine corresponding knowledge information；The equipment 1 will in step s3 The knowledge information is provided to the user.

Specifically, the equipment 1 identification and user's submission in reference to problem base in step sl The corresponding object reference problem of primal problem.The primal problem that the user submits is i.e. for example online The problem of user inputs in the class intelligent Answer System of customer service robot etc. one, for example, in Alibaba Taobao's shopping platform putd question to：" when Yuebao arrives account if being produced", " how to obtain panning Coin" the problems such as.The reference problem base is a typical problem storehouse of one class customer problem of correspondence, For example " when Yuebao arrives account if being produced", " Yuebao produce arrive the account time", " Yuebao turns Go out to when deposit card can arrive account", the example above energy merger is a typical problem, That is " Yuebao produces the account time ".It is herein with reference to asking in this typical problem storehouse and typical problem Exam pool and a citing saying with reference to problem, the original that the reference problem base can be submitted to user The historical data of beginning problem was entered screening or removed after noise data to be clustered using clustering algorithm, from And be a typical problem storehouse by some Similar Problems cluster.Further, it is also possible to set up mark using artificial The method of quasi- problem base, however it is not limited to which this is several, it is other to service number to directly use historical problem It is should be included according to the typical problem storehouse set up in the protection domain of the application.

Wherein, can be by recognizing in keyword or subsumption problem after getting the problem of user submits The method such as validity feature and random forest, Recognition with Recurrent Neural Network scheduling algorithm by customer problem and standard Problem is matched, so that by standard corresponding with the primal problem that user is submitted in typical problem storehouse Problem identification comes out, and so the processing procedure with identification function can be such as the typical problem in Fig. 7 Identification model.The typical problem identified can be one or more, these are identified and user The corresponding typical problem of primal problem is the object reference problem.Multiple typical problems can be one The scaling problem of individual central issue, or similar candidate criteria problem, can be matched only as needed One or multiple specific requirements by historical data analysis confidence level highest or user being evaluated Confidence point highest typical problem preferably comes out.

Those skilled in the art will be understood that above-mentioned foundation with reference to problem base and from reference in problem base The mode of identification object reference problem is only for example, other foundation ginsengs that are existing or being likely to occur from now on Examine exam pool and be such as applicable to this Shen from the mode with reference to identification object reference problem in problem base Please, it should also be included within the application protection domain, and be incorporated herein by reference herein.

Then, the equipment 1 maps mould according to the object reference problem and question and answer in step s 2 Type determines corresponding knowledge information.The knowledge information is corresponding with the object reference problem One or more knowledge points or corresponding solution, such as article or yourself kit, yourself kit Can be the through link for the functions such as a product is opened, closed, bill is checked, these knowledge letter Breath can use knowledge/abstract generation module to be carried in embodiment, and abstract generation module realizes text The additions and deletions of chapter or knowledge point, which change, the operation such as looks into.Each object reference problem is matched and corresponding known Information is known to be supplied to user to carry out the function refering to as described question and answer mapping model, and such as article leads to Question and answer mapping block carry is crossed under some typical problem, if according to user's primal problem and the standard Problem can find this article by question and answer mapping block.

Those skilled in the art will be understood that the management of above-mentioned correspondence knowledge information and map mould by question and answer The mode of the mapping of block is only for example, other existing or corresponding knowledge informations for being likely to occur from now on Manage and the application is such as applicable to by the mode of the mapping of question and answer mapping block, should also be included in this Shen Please be within protection domain, and be incorporated herein by reference herein.

Then, the equipment 1 provides the knowledge information to the user in step s3.I.e. By intelligent Answer System, such as online customer service robot is by corresponding article, knowledge point, from assistant engineer The information of tool is supplied to client by the visible mode of user so that user obtains the proposed original of correspondence The answer of beginning problem.Preferably, can be according to oneself requirement to institute after user's acquisition knowledge information The knowledge information is provided to be evaluated, or intelligent Answer System based on user by evaluating anti- Feedback, the behavior for for example clicking on knowledge point by user judges whether output is accurate, optimizes, and supplements The mapping relations of question and answer mapping block, so as to be more suitable for user's by what the knowledge point fed back was adjusted Currently proposed the problem of.

Preferably, in step s 2 the equipment 1 according to the object reference problem and the user User state information determine corresponding knowledge information in the question and answer mapping model.User's shape State information refers to the individual difference informations such as the account status of user, Entered state, such as in Alibaba Taobao's commodity and service platform in account just have two kinds of personal account and Merchant Account, both account shapes State is a kind of user state information.Because the corresponding knowledge information more than one of object reference problem, So needing to refer to user state information to determine the active user for being best suitable for proposing user's primal problem Required knowledge information is that common question and answer mapping is one-to-one relation, but for complicated industry Business is one-to-many relation.Payment cipher is for example reset, is different for personal user and businessman user Solution, therefore to distinguish.And reset payment cipher this is a typical problem, therefore Question and answer can have two schemes when mapping, one kind is two policing rules of configuration：rule1：Account Type=individual, exports knowledge point ID1, and correspondence personal user resets the operating method of payment cipher； Rule2：Account Type=businessman, exports knowledge point ID2, and corresponding merchant user resets payment cipher Operating method.Also a kind of is exactly to be showed by the form set, two leaves of one tree, respectively Knowledge point ID1 and ID2 are loaded, the corresponding condition of each leaf is exactly Account Type=individual, account Type=businessman.By causing the knowledge information of be presented to user more to fit user's with upper type Personal touch so that the question and answer accuracy rate lifting of intelligent Answer System, while also lifting Consumer's Experience.

Those skilled in the art will be understood that above-mentioned for object reference problem combination user state information Determine that the mode of knowledge information is only for example in question and answer mapping model, other are existing or from now on may The management of the corresponding knowledge information of appearance and by the mapping of question and answer mapping block be directed to object reference problem Determine that the mode of knowledge information is such as applicable to this in question and answer mapping model with reference to user state information Application, should also be included within the application protection domain, and be incorporated herein by reference herein.

It is highly preferred that the user state information includes following at least any one：The application of the user Access behavior；The Account Type information of the user；The account status information of the user.Wherein, The application access behavior of the user refers to that the history that user is recorded in systems accesses behavioral data, Such as software where access time or user uses life cycle phase, in another example user is daily to business Type of access preference of product etc..The Account Type information of the user refers to user account feature, for example User as exemplified above and the Account Type of businessman；The user state information refers to other not most All individual difference informations that can be used to judge customer problem demand difference.

Preferably, methods described also includes step S4 (not shown), in step s 4 the equipment 1 The reference problem base is set up by carrying out clustering processing to a plurality of problem log, wherein, the reference Problem base includes multiple problems cluster, and each problem cluster includes a plurality of described problem and recorded and correspondingly Reference problem.Described problem record refers to that historic customer personnel remember with the service in user interaction process Record, or the historical record that question answering system before is recorded.The clustering processing refers to by poly- Some problem is recorded representative class issue handling formation and refers to problem by the processing of class algorithm.It is described Problem cluster be some problem representated by the problem of type and with the reference corresponding to this problem types The problem cluster of problem, i.e., one solves a type of problem.

Wherein, the method for the reference problem base is set up including the use of hierarchical clustering side by clustering processing The problem of method is to giving log data set carries out the decomposition of level, untill certain condition is met.I.e. Using bottom-up strategy, each object is then combined with these clusters as a cluster first For increasing cluster, until all objects are all in a cluster, or some finish condition is expired Foot, most hierarchy clustering methods belong to this class, and they are simply between cluster in the definition of similarity It is different, therefore clustering processing method in the application is not limited to cited mode.Specifically, Provide the N number of problem log object to be clustered and N*N distance matrix, or similitude square Battle array also can, here, N numerical value is dynamic, according to user to the proposition frequency of problem and problem Quantity etc. is determined, for example, can be configured between 10-100.The Hierarchical cluster side illustrated herein The basic step of method is as follows：First, each problem log object is classified as a class, N classes is obtained, Per class only comprising an object.Then the distance between class and class are calculated, that is, they are included The distance between object, here, apart from being exactly similarity between similarity, two customer problems, Text matches algorithm can be used to calculate.Then, two classes of the score value of similarity recently are found to be closed And, then total class number has lacked one.Then new class is recalculated and the distance between class of haveing been friends in the past. Repeat distance to calculate with merging all kinds of conjunctions of N representated by the last initial N problem logs And into untill a class, i.e., contain N number of object in a class here.Further, with reference to asking The generation of exam pool is optimized for step of illustrating as follows herein in connection with clustering processing：According to data mining client on line Primal problem or problem log and clustering processing；Algorithm output cluster data is supplied to knowledge information to store Management module such as knowledge base；Knowledge base platform to reference to problem base carry out review operations, for example based on The original customer problem of the threshold value discard portion of similarity, set threshold value can between 0.1-0.9, Or indirect labor is audited；Knowledge base have recorded all examination ＆ verification storages, ignore not by examination ＆ verification The problem of the peration data such as record data, and feed back to cluster processing algorithm；Cluster processing algorithm is carried out Optimization, the algorithm after optimization carries out data mining and clustering processing again.Thus formed reference problem Storehouse covers all user's ways to put questions, compensate for traditional conversion by being artificially formed optimization problem list The shortcoming that time is slow, efficiency is low, user's way to put questions is clustered onto a typical problem, so as to lift effect Rate.

Those skilled in the art will be understood that above-mentioned clustering processing and feedback optimized mode are only act Example, other existing or clustering processings for being likely to occur from now on and feedback optimized mode are for example applicable In the application, it should also be included within the application protection domain, and be incorporated herein by reference herein.

It is highly preferred that described problem record meets predetermined record screening conditions.Because described problem is remembered Record is historic customer personnel and the service log in user interaction process, or question answering system before The historical record recorded, therefore inaccurate record or the poor record of user's evaluation is had, can To set such as evaluation star more than four stars or be cited as accuracy rate higher than 80 percent, and or comment Valency is high, handled without the poor service log for commenting word.Here, the record screening conditions are pair The condition of accuracy rate or confidence rate set by history question and answer service log, so that for forming ginseng The data for examining exam pool reduce the low data being unsuitable for for user feedback of quality, lift accuracy rate.

Those skilled in the art will be understood that the mode of the predetermined record screening conditions of above-mentioned setting is only Citing, the mode of the predetermined record screening conditions of other settings that are existing or being likely to occur from now on such as may be used Suitable for the application, it should also be included within the application protection domain, and include by reference herein In this.

Fig. 2 shows the method flow diagram of the step S1 according to one preferred embodiment of the application.It is described Step S1 includes step S11, step S12, step S13, step S14.

Wherein, the primal problem that the equipment 1 is determined with user submits in step s 11 is corresponding Problem characteristic information；The equipment 1 is asked according to described problem characteristic information in reference in step s 12 The object reference problem corresponding with the primal problem is recognized in exam pool；It is described in step s 13 to set Standby 1 recognizes the one or more candidates corresponding with the primal problem that user submits in reference to problem base Object reference problem；The equipment 1 described in step S14 is according to the candidate target with reference to problem correspondence Confidence information from one or more of candidate targets refer to problem in selected objective target refer to problem.

Specifically, the primal problem that the equipment 1 is determined with user submits in step s 11 is corresponding The problem of characteristic information.The primal problem that the user submits is i.e. one such as online customer service robot The problem of user inputs in class intelligent Answer System, for example, in Taobao's shopping platform of Alibaba Putd question to：" when Yuebao arrives account if being produced”.Described corresponding the problem of characteristic information be exemplified by Such as feature category of keyword or other representative users extracted according to user's enquirement or state data Know.For example, the primal problem of user is " afternoon 15 January 8:00 Yuebao produced is properly arrived Up to the savings account of binding", keyword " Yuebao " " producing " is extracted the problem of such and " arrives account The information such as time " " deposit card " are corresponding problem characteristic information.Extract described problem feature letter Breath make it that the matching of object reference problem is more accurate, causes so as to obtain accurately correspondence knowledge information The accuracy rate lifting of intelligent answer.

Preferably, the equipment 1 is submitted according to user in step s 11 primal problem and the use The user state information at family determines corresponding problem characteristic information.The user state information is used as described Family status information refers to the individual difference informations such as the account status of user, Entered state, therefore can use To extract corresponding feature so that it is determined that corresponding problem characteristic information, for example, the problem of user yes Problem is that " which member's rights and interests have" belong to and determine according to the membership in the state of user The enquirement of user is the member of that service field, e.g. Taobao member or day cat member's rights and interests, from And orientation problem is characterized in belong to Taobao or day cat.

Those skilled in the art will be understood that the mode of above-mentioned determination problem characteristic information is only for example, its He is such as applicable to the application at the mode of determination problem characteristic information that is existing or being likely to occur from now on, Also it should be included within the application protection domain, and be incorporated herein by reference herein.

Then, the equipment 1 is referring to problem base according to described problem characteristic information in step s 12 The middle identification object reference problem corresponding with the primal problem.The identification and the primal problem Corresponding object reference problem, as problem identification, are the problem of a kind of customer problem are classified, point Class is come according to the seek help data of enquirement of the data such as the user behavior that gathers in real time or account and user The class mark of user characteristics is portrayed, and then the classification situation of Future Data can be predicted.Can in embodiment This function can be carried out by problem identification module, be to be mapped to typical problem the problem of user is putd question to One in storehouse.Specifically, problem identification algorithm includes random forests algorithm or RNN (Recurrent Neural Network, Recognition with Recurrent Neural Network), this is a kind of artificial neuron of node orientation connection cyclization Network.The internal state of this network can show dynamic time sequence behavior.RNN can be using inside it Memory handle the list entries of arbitrary sequence, this allows it to be easier processing such as the hand not being segmented Write identification, speech recognition etc..Such as connect exemplified above, " Yuebao is navigated to after obtaining keyword feature Produce account typical problem ", then by question and answer mapping block below, extract " 15:00 " knowledge letter Breath, it is corresponding two hours to find out typical problem, provides " 17:The user feedback of account is arrived before 00 " point.

Those skilled in the art will be understood that the above-mentioned identification target ginseng corresponding with the primal problem The mode of examination topic is only for example, and other existing or identifications for being likely to occur from now on original are asked with described The mode of the corresponding object reference problem of topic is such as applicable to the application, should also be included in the application and protect Protect within scope, and be incorporated herein by reference herein.

Then, the equipment 1 recognizes the original submitted with user in reference to problem base in step s 13 The corresponding one or more candidate targets of beginning problem refer to problem.Calculated by corresponding problem identification The object reference problem that method is drawn can be one or more, because finally feeding back to knowing for user It is that can just cause corresponding knowledge information accurate based on a unique object reference problem to know information optimal Really and with specific aim.Therefore, the candidate target with reference to problem be foregoing drawn one or More than one object reference problem is in preferably unique preceding state.For example, the description of user is passed through Clustering algorithm generates a typical problem and many scaling problems, such as in the service platform of Alipay User puts question to：" why make to have more than is needed flower", " flower how not open" " flower can not swash It is living", " open flower have what condition" these customer problems belong to a typical problem： Flower condition is opened, these problems are all the scaling problems as this typical problem, these extensions I.e. be likely to become candidate target refers to problem to problem during identification.

Then, the equipment 1 described in step S14 is put with reference to problem is corresponding according to the candidate target Confidence information refers to problem from one or more of candidate targets with reference to selected objective target in problem.It is described Confidence level is to pass through evaluation or fraction of the user to candidate target with reference to problem in historical data, be can be used for Candidate target is preferred with reference to problem.For example, the enquirement to user carries out problem identification, it is necessary to match To a unique typical problem, if matching multiple typical problems, output confidence point highest One, wherein, confidence point is a score value of problem identification inside modules algorithm, and score value shows more greatly Confidence point is higher.Based on this so that the accuracy lifting of the object reference problem selected to user.

Those skilled in the art will be understood that above-mentioned selected objective target is only for example with reference to the mode of problem, its He is such as applicable to the application by selected objective target that is existing or being likely to occur from now on reference to the mode of problem, Also it should be included within the application protection domain, and be incorporated herein by reference herein.

Fig. 3 shows a kind of method flow for being used to provide knowledge information according to the application other side Figure.Methods described includes step S5, step S1, step S2, step S3.

Wherein, whether the primal problem that the detection of equipment 1 user submits in step s 5 is semantic complete It is whole；If the equipment 1 primal problem is semantic complete in step sl, in reference to problem base Recognize the corresponding object reference problem of the primal problem；The equipment 1 is according to institute in step s 2 State object reference problem and question and answer mapping model determines corresponding knowledge information；It is described in step s3 Equipment 1 provides the knowledge information to the user.

Here, step S2 described in Fig. 1, step S3 are identical with step S2, step S3 in Fig. 3 Or it is similar, it will not be repeated here.

Specifically, whether the primal problem that the detection of equipment 1 user submits in step s 5 is semantic Completely.Judge whether the whether complete namely problem of user semantic is complete, it is closely related with business scenario, For example, " when Yuebao arrives account if being produced" this user put question to just lack the time of transferring accounts, gold Volume, it is transferred to the information such as bank.Therefore need to define miscellaneous service scene, operation personnel can be passed through initial stage Original definition is carried out to the semantic of miscellaneous service scene, the later stage can be optimized by data feedback, example The semantic key words in high question and answer are such as evaluated user to be extracted and then mark some after being integrated Vocabulary in the corresponding primal problem of quasi- problem is calculated, and accounting is higher and be not auxiliary word or the tone The vocabulary of word is fed back.

Preferably, the equipment 1 obtains user based on the increasing of semantic supplement prompt message in step s 5 Plus on the primal problem the problem of information, and the primal problem is updated, until after updating The primal problem is semantic complete.Even user puts question to semantic imperfect, then asks in reply user, Zhi Daoshang Hereafter untill semantic completion, example of such as passing the imperial examinations at the provincial level above, by user put question to lacking in transfer accounts the time, the amount of money, It is transferred to the information such as bank and carries out prompting completion, until the related equal completion of keyword carries out mesh below again Related work of the mark with reference to problem.Cause the original inquiry of user more by prompting and completion repeatedly With specific aim, and certain user state information can be collected into, so as to be conducive to corresponding phase Knowledge information is closed to be judged.

Those skilled in the art will be understood that the mode of above-mentioned Semantic detection is only for example, and other are existing Or the mode for the Semantic detection being likely to occur from now on is such as applicable to the application, the application should be also included in Within protection domain, and it is incorporated herein by reference herein.

Then, if the equipment 1 primal problem is semantic complete in step sl, asked in reference The corresponding object reference problem of the primal problem is recognized in exam pool.I.e. the user carried it is original On the basis of problem semanteme is complete, system can be used for extracting feature or using other matching ways to referring to Reference problem in problem base is matched, know method for distinguishing with it is described previously identical.Based on complete Semanteme is so that the feature or the information of reference that system can be extracted are more accurate and comprehensive, so as to avoid Because the matching of object reference problem is inaccurate caused by the missing of customer problem semanteme or keyword, and then Lift the accuracy of intelligent answer.

Fig. 4 shows a kind of equipment signal for being used to provide knowledge information according to the application other side Figure.The equipment 1 includes object reference problem identification device 11, question and answer mapping device 12, knowledge Managing device 13 is provided.

Wherein, the identification in reference to problem base of object reference problem identification device 11 is submitted with user The corresponding object reference problem of primal problem；Question and answer mapping device 12 is asked according to the object reference Topic and question and answer mapping model determine corresponding knowledge information；Knowledge provides managing device 13 and known described Know information to provide to the user.

Specifically, the identification in reference to problem base of object reference problem identification device 11 is submitted with user The corresponding object reference problem of primal problem.The primal problem that the user submits for example is existing The problem of user inputs in the class intelligent Answer System of line customer service robot etc. one, for example, in Arriba Taobao's shopping platform of bar is putd question to：" when Yuebao arrives account if being produced", " how to obtain panning Coin" the problems such as.The reference problem base is a typical problem storehouse of one class customer problem of correspondence, For example " when Yuebao arrives account if being produced", " Yuebao produce arrive the account time", " Yuebao turns Go out to when deposit card can arrive account", the example above energy merger is a typical problem, That is " Yuebao produces the account time ".It is herein with reference to asking in this typical problem storehouse and typical problem Exam pool and a citing saying with reference to problem, the original that the reference problem base can be submitted to user The historical data of beginning problem was entered screening or removed after noise data to be clustered using clustering algorithm, from And be a typical problem storehouse by some Similar Problems cluster.Further, it is also possible to set up mark using artificial The method of quasi- problem base, however it is not limited to which this is several, it is other to service number to directly use historical problem It is should be included according to the typical problem storehouse set up in the protection domain of the application.

Then, question and answer mapping device 12 is determined according to the object reference problem and question and answer mapping model Corresponding knowledge information.The knowledge information be corresponding with the object reference problem one or Multiple knowledge points or corresponding solution, such as article or yourself kit, yourself kit can be It is the through link for the functions such as a product is opened, closed, bill is checked, these knowledge informations are in reality Applying can use knowledge/abstract generation module to be carried in example, abstract generation module is realized article or known Know point additions and deletions change look into wait operate.Each object reference problem is matched into corresponding knowledge information To be supplied to user to carry out the function refering to as described question and answer mapping model, such as article passes through question and answer Mapping block carry is under some typical problem, if being with the typical problem according to user's primal problem This article can be found by question and answer mapping block.

Then, knowledge provides managing device 13 and provides the knowledge information to the user.Lead to Cross intelligent Answer System, such as online customer service robot is by corresponding article, knowledge point, yourself kit Information client is supplied to by the visible mode of user so that user obtain correspondence proposed it is original The answer of problem.Preferably, can be according to oneself requirement to being carried after user's acquisition knowledge information Evaluated for the knowledge information, or intelligent Answer System is fed back by being evaluated based on user, The behavior for for example clicking on knowledge point by user judges whether output is accurate, optimizes, and supplements question and answer The mapping relations of mapping block, so as to be more suitable for the current of user by what the knowledge point fed back was adjusted The problem of proposed.

Preferably, the question and answer mapping device 12 is according to the object reference problem and the user User state information determines corresponding knowledge information in the question and answer mapping model.The User Status Information refers to the individual difference informations such as the account status of user, Entered state, such as in Alibaba Account just has two kinds of personal account and Merchant Account, both account status in Taobao's commodity and service platform As a kind of user state information.Because the corresponding knowledge information more than one of object reference problem, institute The active user institute for being best suitable for proposing user's primal problem is determined to need to refer to user state information The knowledge information needed is that common question and answer mapping is one-to-one relation, but for complicated business It is one-to-many relation.Payment cipher is for example reset, is different for personal user and businessman user Solution, therefore to distinguish.And reset payment cipher this is a typical problem, therefore asking There can be two schemes when answering mapping, one kind is two policing rules of configuration：rule1：Account class Type=individual, exports knowledge point ID1, and correspondence personal user resets the operating method of payment cipher；Rule2： Account Type=businessman, exports knowledge point ID2, and corresponding merchant user resets the operation side of payment cipher Method.Also a kind of is exactly to be showed by the form set, and two leaves of one tree, respectively loading is known Know point ID1 and ID2, the corresponding condition of each leaf is exactly Account Type=individual, Account Type=business Family.By caused with upper type the knowledge information of be presented to user more fit user individual it is special Point so that the question and answer accuracy rate lifting of intelligent Answer System, while also lifting Consumer's Experience.

Preferably, the equipment 1 is also included with reference to the (not shown) of problem base generating means 14, described With reference to problem base generating means 14 reference is set up by carrying out clustering processing to a plurality of problem log Problem base, wherein, the reference problem base includes multiple problem clusters, and each problem cluster includes A plurality of described problem record and corresponding reference problem.Described problem record refer to historic customer personnel with Service log in user interaction process, or the historical record that question answering system before is recorded. The clustering processing refers to that some problem is recorded into a representative class by the processing of clustering algorithm asks Topic processing, which is formed, refers to problem.Described problem cluster be some problem representated by the problem of type and with Reference problem corresponding to this problem types, i.e., one problem cluster solves a type of problem.

Fig. 5 shows setting according to the object reference problem identification device of the application another preferred embodiment Standby schematic diagram.The object reference problem identification device 11 includes problem characteristic determining unit 111, mesh Mark is joined with reference to problem determining unit 112, candidate target with reference to problem determining unit 113, selected objective target Determining unit 114 is inscribed in examination.

Wherein, problem characteristic determining unit 111 determines corresponding to ask with the primal problem of user's submission Inscribe characteristic information；Object reference problem determining unit 112 is asked according to described problem characteristic information in reference The object reference problem corresponding with the primal problem is recognized in exam pool；Candidate target is true with reference to problem Order member 113 recognizes one or many corresponding with the primal problem that user submits in reference to problem base Individual candidate target refers to problem；Selected objective target is with reference to problem determining unit 114 according to the candidate target With reference to the corresponding confidence information of problem from one or more of candidate targets with reference to preferred in problem Object reference problem.

Specifically, the primal problem that problem characteristic determining unit 111 is determined with user submits is corresponding Problem characteristic information.The primal problem that the user submits is i.e. in a class such as online customer service robot The problem of user inputs in intelligent Answer System, for example, Taobao's shopping platform in Alibaba enters Row is putd question to：" when Yuebao arrives account if being produced”.Described corresponding the problem of, characteristic information was for example The feature class of keyword or other representative users extracted according to data such as user's enquirement or states are identified. For example, the primal problem of user is " afternoon 15 January 8:00 Yuebao produced is properly reached and tied up Fixed savings account", keyword " Yuebao " " producing " " arriving the account time " is extracted the problem of such Information such as " deposit cards " is corresponding problem characteristic information.Described problem characteristic information is extracted to cause The matching of object reference problem is more accurate, so as to obtain accurately correspondence knowledge information so that intelligence is asked The accuracy rate lifting answered.

Preferably, described problem characteristics determining unit 111 is submitted according to user primal problem and described The user state information of user determines corresponding problem characteristic information.The user state information is as described User state information refers to the individual difference informations such as the account status of user, Entered state, therefore can be with For extracting corresponding feature so that it is determined that corresponding problem characteristic information, for example, being the problem of user The problem of be that " which member's rights and interests have" belonged to and true according to the membership in the state of user The enquirement for determining user is the member of that service field, e.g. Taobao member or day cat member's rights and interests, So as to which orientation problem is characterized in belong to Taobao or day cat.

Then, object reference problem determining unit 112 is referring to problem according to described problem characteristic information The object reference problem corresponding with the primal problem is recognized in storehouse.The identification original is asked with described The corresponding object reference problem of topic, as problem identification, are the problem of a kind of customer problem are classified, Classification is the data of enquirement of being sought help according to the data such as the user behavior that gathers in real time or account and user To portray the class mark of user characteristics, and then the classification situation of Future Data can be predicted.In embodiment It is to be mapped to standard the problem of user is putd question to ask but this function can be carried out by problem identification module One in exam pool.Specifically, problem identification algorithm includes random forests algorithm or RNN (Recurrent Neural Network, Recognition with Recurrent Neural Network), this is a kind of artificial neuron of node orientation connection cyclization Network.The internal state of this network can show dynamic time sequence behavior.RNN can be using inside it Memory handle the list entries of arbitrary sequence, this allows it to be easier processing such as the hand not being segmented Write identification, speech recognition etc..Such as connect exemplified above, " Yuebao is navigated to after obtaining keyword feature Produce account typical problem ", then by question and answer mapping block below, extract " 15:00 " knowledge letter Breath, it is corresponding two hours to find out typical problem, provides " 17:The user feedback of account is arrived before 00 " point.

Then, candidate target is recognized in reference to problem base with reference to problem determining unit 113 and carried with user One or more candidate targets that the primal problem of friendship is corresponding refer to problem.I.e. by it is corresponding the problem of The object reference problem that recognizer is drawn can be one or more, because finally feeding back to use The optimal knowledge information at family is just to cause corresponding knowledge based on a unique object reference problem Information is accurate and with specific aim.Therefore, the candidate target is foregoing drawn with reference to problem One or more object reference problem is in preferably unique preceding state.For example, retouching user State and a typical problem and many scaling problems, such as service in Alipay are generated by clustering algorithm User puts question in platform：" why make to have more than is needed flower", " flower how not open" " flower It can not activate", " open flower have what condition" these customer problems belong to a standard Problem：Flower condition is opened, these problems are all the scaling problems as this typical problem, this I.e. be likely to become candidate target refers to problem to a little scaling problems during identification.

Then, selected objective target refers to problem pair with reference to problem determining unit 114 according to the candidate target The confidence information answered is asked from one or more of candidate targets with reference to selected objective target reference in problem Topic.The confidence level is to pass through user in historical data to refer to candidate target the evaluation of problem or divide Number, available for candidate target with reference to the preferred of problem.For example, the enquirement to user carries out problem identification, Need to match a unique typical problem, if matching multiple typical problems, export confidence Divide highest one, wherein, confidence point is a score value of problem identification inside modules algorithm, score value Show that confidence point is higher more greatly.Based on this so that the accuracy of the object reference problem selected to user Lifting.

Fig. 6 shows a kind of equipment signal for being used to provide knowledge information according to the application other side Figure.The equipment 1 includes primal problem Semantic detection device 25, object reference problem identification device 21st, question and answer mapping device 22, knowledge provide managing device 23.

Wherein, whether the primal problem that the detection of primal problem Semantic detection device 25 user submits is semantic Completely；If the primal problem of object reference problem identification device 21 is semantic complete, problem is being referred to The corresponding object reference problem of the primal problem is recognized in storehouse；Question and answer mapping device 22 is according to described Object reference problem and question and answer mapping model determine corresponding knowledge information；Knowledge provides managing device 23 The knowledge information is provided to the user.

Here, question and answer mapping device described in Fig. 4 12, knowledge are provided in managing device 13 and Fig. 6 The question and answer mapping device 22, knowledge offer managing device 23 are same or similar, will not be repeated here.

Specifically, primal problem Semantic detection device 25 detection user submit primal problem whether language It is adopted complete.Judge whether the whether complete namely problem of user semantic is complete, and the close phase of business scenario Close, for example, " when Yuebao arrives account if being produced" this user put question to just lack the time of transferring accounts, The amount of money, it is transferred to the information such as bank.Therefore need to define miscellaneous service scene, initial stage can be by runing people Member carries out original definition to the semantic of miscellaneous service scene, and the later stage can be optimized by data feedback, For example user is evaluated some after the semantic key words in high question and answer are extracted and then integrated Vocabulary in the corresponding primal problem of typical problem is calculated, and accounting is higher and be not auxiliary word or language The vocabulary of gas word is fed back.

Preferably, the primal problem Semantic detection device 25 is obtained user and pointed out based on semantic supplement Information it is increased on the primal problem the problem of information, and the primal problem is updated, until more The primal problem after new is semantic complete.Even user puts question to semantic imperfect, then asks in reply user, Untill the semantic completion of context, example of such as passing the imperial examinations at the provincial level above, by user put question to lacking in transfer accounts the time, The amount of money, it is transferred to the information such as bank and carries out prompting completion, until after the related equal completion of keyword is carried out again The related work of the object reference problem in face.By prompting and completion repeatedly the original of user is carried Asking more has specific aim, and can be collected into certain user state information, so as to be conducive to right The relevant knowledge information answered is judged.

Then, if the primal problem of object reference problem identification device 21 is semantic complete, in reference The corresponding object reference problem of the primal problem is recognized in problem base.The original carried in the user On the basis of beginning problem semanteme is complete, system can be used for extracting feature or using other matching ways to joining Examination exam pool in reference problem matched, knowledge method for distinguishing with it is described previously identical.Based on complete Semanteme make it that feature that system can extract or the information of reference are more accurate and comprehensively, so as to keep away Exempt from, because the matching of object reference problem is inaccurate caused by the missing of customer problem semanteme or keyword, to enter And lift the accuracy of intelligent answer.

Preferably, Fig. 7 is a kind of robot online chatting question and answer matching process using application scheme, It is a kind of typical problem storehouse and its question and answer matching system, including：Typical problem cluster module, problem are known Other module, question and answer mapping block, abstract generation module.The corresponding target ginseng of concrete application such as user Examination is entitled:Reset password.Question and answer mapping block according to user state information for example：Password type " branch Pay, log in "；Account Type " personal, businessman " is progressive to find a unique knowledge point.

Wherein, the excavation first to customer problem in service log generates typical problem storehouse, then step It is as follows：Step A：Robot judged the original way to put questions of user, judge to put question to it is semantic whether Completely；Step B：If puing question to semantic imperfect, user is asked in reply, until the semantic completion of context Untill, into step D；Step C：If puing question to semantic complete, into step D；Step D： Problem identification is carried out, a unique typical problem is matched, if matching multiple typical problems, Export confidence point highest one；Step E：Robot is reflected based on the corresponding question and answer of this typical problem Penetrate module and disposably obtain the information such as user behavior, account status, matched based on these status informations A unique knowledge point ID；Step F：Knowledge point ID is exported to user.It is highly preferred that base The behavior that feedback is user's click knowledge point is evaluated in user and judges whether output is accurate, and question and answer are mapped Module is optimized and supplemented.So as to carry out problem to the answer in service log, knowledge base is driven Update, and the mapping relations of problem answers are configured by decision tree, it is accurate that lifting question and answer map Rate, final knowledge point faces customer problem directly and User Status is more friendly.

Obviously, those skilled in the art can to the application carry out it is various change and modification without departing from Spirit and scope.So, if these modifications and variations of the application belong to the application power Within the scope of profit requirement and its equivalent technologies, then the application is also intended to exist comprising these changes and modification It is interior.

It should be noted that the application can be carried out in the assembly of software and/or software and hardware, For example, can be using application specific integrated circuit (ASIC), general purpose computer or any other is similar hard Part equipment is realized.In one embodiment, the software program of the application can pass through computing device To realize steps described above or function.Similarly, the software program of the application (includes the number of correlation According to structure) it can be stored in computer readable recording medium storing program for performing, for example, RAM memory, magnetic Or CD-ROM driver or floppy disc and similar devices.In addition, some steps or function of the application can be used Hardware realizes, for example, as coordinating with processor so as to performing the circuit of each step or function.

In addition, the part of the application can be applied to computer program product, such as computer program Instruction, when it is computer-executed, by the operation of the computer, can call or provide basis The present processes and/or technical scheme.And the programmed instruction of the present processes is called, it may be deposited Store up in fixed or moveable recording medium, and/or by broadcast or other signal bearing medias Data flow and be transmitted, and/or be stored according to the computer equipment of described program instruction operation In working storage.Here, including a device, the device bag according to one embodiment of the application The memory for storing computer program instructions and the processor for execute program instructions are included, its In, when the computer program instructions are by the computing device, trigger the plant running and be based on foregoing According to the methods and/or techniques scheme of multiple embodiments of the application.

It is obvious to a person skilled in the art that the application is not limited to the thin of above-mentioned one exemplary embodiment Section, and in the case of without departing substantially from spirit herein or essential characteristic, can be with other specific Form realizes the application.Therefore, no matter from the point of view of which point, embodiment all should be regarded as exemplary , and be nonrestrictive, scope of the present application is limited by appended claims rather than described above It is fixed, it is intended that all changes fallen in the implication and scope of the equivalency of claim are included In the application.The right that any reference in claim should not be considered as involved by limitation will Ask.Furthermore, it is to be understood that the word of " comprising " one is not excluded for other units or step, odd number is not excluded for plural number. The multiple units or device stated in device claim can also be passed through soft by a unit or device Part or hardware are realized.The first, the second grade word is used for representing title, and is not offered as any spy Fixed order.

Claims

1. a kind of method for providing knowledge information, including：

The knowledge information is provided to the user.

2. according to the method described in claim 1, wherein, methods described also includes：

The reference problem base is set up by carrying out clustering processing to a plurality of problem log, wherein, the ginseng Examination exam pool includes multiple problems cluster, and each problem cluster includes a plurality of described problem and recorded and correspondingly Reference problem.

3. method according to claim 2, wherein, described problem record meets predetermined record sieve Select condition.

4. according to the method described in claim 1, wherein, it is described according to the object reference problem and to ask Answer mapping model and determine that corresponding knowledge information includes：

According to the user state information of the object reference problem and the user in the question and answer mapping model It is middle to determine corresponding knowledge information.

5. method according to claim 4, wherein, the user state information include it is following at least Any one：

The application access behavior of the user；

The Account Type information of the user；

The account status information of the user.

6. according to the method described in claim 1, wherein, it is described with reference to identification and user in problem base The object reference problem that the primal problem of submission is corresponding includes：

It is determined that with user submit primal problem it is corresponding the problem of characteristic information；

The mesh corresponding with the primal problem is recognized in reference to problem base according to described problem characteristic information Mark refers to problem.

7. method according to claim 6, wherein, the primal problem that the determination is submitted with user Corresponding the problem of, characteristic information included：

The problem of primal problem and the user state information of the user submitted according to user determine correspondence is special Reference ceases.

8. according to the method described in claim 1, wherein, it is described with reference to identification and user in problem base The object reference problem that the primal problem of submission is corresponding includes：

The identification one or more candidate mesh corresponding with the primal problem that user submits in reference to problem base Mark refers to problem；

The corresponding confidence information of problem is referred to from one or more of candidate's mesh according to the candidate target Mark refers to problem with reference to selected objective target in problem.

9. according to the method described in claim 1, wherein, methods described also includes：

Detect that whether semantic the primal problem that user submits is complete；

Wherein, the identification target corresponding with the primal problem that user submits in reference to problem base is joined Examination topic includes：

If the primal problem is semantic complete, the corresponding mesh of the primal problem is recognized in reference to problem base Mark refers to problem.

10. method according to claim 9, wherein, the primal problem that the detection user submits Whether semanteme completely also includes：

Obtain user be based on semantic supplement prompt message it is increased on the primal problem the problem of information, And the primal problem is updated, until the primal problem after updating is semantic complete.

11. a kind of equipment for providing knowledge information, including：

Object reference problem identification device, original the asking for identification and user's submission in reference to problem base The corresponding object reference problem of topic；

Question and answer mapping device, it is corresponding for being determined according to the object reference problem and question and answer mapping model Knowledge information；

12. equipment according to claim 11, wherein, the equipment also includes：

With reference to problem base generating means, for setting up described by carrying out clustering processing to a plurality of problem log With reference to problem base, wherein, the reference problem base includes multiple problem clusters, each problem cluster bag Include a plurality of described problem record and corresponding reference problem.

13. equipment according to claim 12, wherein, described problem record meets predetermined record Screening conditions.

14. equipment according to claim 11, wherein, the question and answer mapping device is used for：

15. equipment according to claim 14, wherein, the user state information is included below extremely It is any one of few：

The application access behavior of the user；

The Account Type information of the user；

The account status information of the user.

16. equipment according to claim 11, wherein, the object reference problem identification device bag Include：

Problem characteristic determining unit, the problem of primal problem for determining with user submits is corresponding feature Information；

Object reference problem determining unit, for being known according to described problem characteristic information in reference to problem base Object reference problem not corresponding with the primal problem.

17. equipment according to claim 16, wherein, described problem characteristics determining unit is used for：

18. equipment according to claim 11, wherein, the object reference problem identification device bag Include：

Candidate target refers to problem determining unit, for the original that identification is submitted with user in reference to problem base The corresponding one or more candidate targets of beginning problem refer to problem；

Selected objective target refers to problem determining unit, and for being referred to according to the candidate target, problem is corresponding to put Confidence information refers to problem from one or more of candidate targets with reference to selected objective target in problem.

19. equipment according to claim 11, wherein, the equipment also includes：

Primal problem Semantic detection device, for detecting that whether semantic the primal problem that user submits is complete；

Wherein, the object reference problem identification device is used for：

20. equipment according to claim 19, wherein, the primal problem Semantic detection device is also For：