Summary of the invention
The objective of the invention is at the deficiencies in the prior art, a kind of question answering system based on man-machine combination and method are provided, introduce artifact, adopt artificial instruction mode, improve self-teaching and the self promotion ability of question answering system, increase hommization and intellectuality simultaneously, guarantee accuracy and the promptness of system.
For achieving the above object, the invention provides a kind of question answering system based on man-machine combination and method, comprising:
Load module is used for input user speech, text and operational order and puts question to information, gathers customer parameter information, and puts question to information and customer parameter information all to be converted to the normative text format information user;
The denoising module is used for text formatting information, carries out denoising and modular structure processing;
Semantic meaning analysis module is used for that user's information of puing question to of denoising and modularization processing is carried out semanteme and resolves;
Message processing module is used for resolving information is handled, and generates result;
The intelligent decision module is for the information processing result being judged and determining;
Output module is used for puing question to answer according to judgement and determination result output user;
Artificial customer service processing module is used for receiving the nothing of exporting and answers and the false answer result, and the pedestrian worker that goes forward side by side handles;
Training module is for receiving artificial result and this intelligent decision module being trained.
Further, this load module comprises:
Voice input module is used for the input of user speech information;
The text load module is used for the input of user version information;
The operation load module is used for the input of user's operational order;
The parameter acquisition module is used for gathering user's various parameter informations;
Text conversion module, being used for puing question to information and customer parameter information translation with the user is text formatting information.
Further, this semantic meaning analysis module comprises:
Word-dividing mode is used for setting up a cover based on the statement of Modern Chinese, the participle algorithm model of sentence structure, utilizes a large amount of basicvocabulary data, adopts smallest particles to divide word algorithm that participle is carried out in natural language and the instruction of user's input;
The weight allocation module is used for to user's linguistic context, context, preference and parameters, carrying out first combination and the weight allocation of key word and intention after the participle, parses key word;
The structured combinations module is used for the key word that parses is carried out structured combinations, obtains analysis result.
Further, this message processing module comprises:
Authentication module is used for judging whether the domain information of this semantic meaning analysis module is comprehensive, if not comprehensive, sets up this context model, preserves this session log, changes corresponding processing platform over to; If directly change corresponding processing platform over to comprehensively;
Processing platform is used for the domain information of checking is handled, and generates the baseline results data.
A kind of answering method based on man-machine combination may further comprise the steps:
Step 1, input user speech, text and operational order are putd question to information, gather customer parameter information;
Step 2, puing question to information and customer parameter information translation with the user is the normative text format information;
Step 3 to text formatting information, is carried out denoising and modular structure processing;
Step 4 is carried out semanteme to user's information of puing question to of denoising and modularization processing and is resolved;
Step 5 is judged and is determined the information processing result;
Step 6 generates the user and puts question to answer, if answer is correct, and directly output; If answer does not have answer or answer is incorrect, submit to artificial customer service to handle problem, judge again and determines according to result, and to judging and deciding ability is trained, export correct option.
Further, this step 4 comprises following substep:
1, adopt the branch word algorithm that participle is carried out in natural language and the instruction of user's input;
2, to user's linguistic context, context, preference and parameters, carry out first combination and the weight allocation of key word and intention, parse key word;
3, the key word that parses is carried out structured combinations, obtain analysis result.
Compared with prior art, the invention has the beneficial effects as follows: by introducing artifact, adopt artificial instruction mode, improved self-teaching and the self promotion ability of question answering system, increase hommization and intellectuality simultaneously, guaranteed accuracy and the promptness of system.
Embodiment
The present invention is described in detail below in conjunction with each embodiment shown in the drawings; but should be noted that; these embodiments are not limitation of the present invention; the function that those of ordinary skills do according to these embodiments, method or structural equivalent transformation or alternative all belong within protection scope of the present invention.
Join shown in Figure 1ly, Fig. 1 is system construction drawing of the present invention.
In the present embodiment, a kind of question answering system based on man-machine combination comprises:
Load module 10 is used for input user speech, text and operational order and puts question to information, gathers customer parameter information, and puts question to information and customer parameter information all to be converted to the normative text format information user;
Denoising module 20 is used for text formatting information, carries out denoising and modular structure processing;
Semantic meaning analysis module 30 is used for that user's information of puing question to of denoising and modularization processing is carried out semanteme and resolves;
Message processing module 40 is used for resolving information is handled, and generates result;
Intelligent decision module 50 is for the information processing result being judged and determining;
This intelligent decision module 50 is one can be learnt, and can sum up the system of conclusion, and it can use to learn and conclude summary by the experiment of given data.In the human perception field of artificial intelligence, we are by can the conduct oneself decision problem of worker's perception aspect of the application of mathematical statistics, and namely by statistical method, this module can similar people equally have simple deciding ability and simple judgement.
Output module 60 is used for puing question to answer according to judgement and determination result output user;
Artificial customer service processing module 70 is used for receiving the nothing of exporting and answers and the false answer result, and the pedestrian worker that goes forward side by side handles;
Training module 80 is for receiving artificial result and this intelligent decision module 50 being trained.
Such as user's input " to five road junctions how by bus Xizhimen ", existing hypothesis can't be made correct answer, and this problem will be pushed to this artificial customer service processing module 70.This artificial customer service processing module 70 is handled this problem, and beams back this training module 80 with specific rule.80 pairs of these intelligent decision modules 50 of this training module are trained, and this intelligent decision module 50 begins this problem is carried out modeling by the correction of training sample, just have afterwards to solve to follow these question marks seemingly or the ability of relevant a series of problems.
Ginseng Fig. 3, Fig. 4 and shown in Figure 5, Fig. 3 is the load module structural drawing; Fig. 4 is the semantic meaning analysis module structural drawing; Fig. 5 is the message processing module structural drawing.
This load module 10 comprises:
Voice input module 101 is used for the input of user speech information;
Text load module 102 is used for the input of user version information;
Operation load module 103 is used for the input of user's operational order;
Parameter acquisition module 104 is used for gathering user's various parameter informations, as time, land used place, terminal device kind and model, network condition, device orientation, speed, acceleration etc.;
Text conversion module 105, being used for puing question to information and customer parameter information translation with the user is text formatting information.
This semantic meaning analysis module 30 comprises:
Word-dividing mode 301 is used for setting up a cover based on the statement of Modern Chinese, the participle algorithm model of sentence structure, utilizes a large amount of basicvocabulary data, adopts smallest particles to divide word algorithm that participle is carried out in natural language and the instruction of user's input.
Weight allocation module 302 is used for to user's linguistic context, context, preference and parameters, carrying out first combination and the weight allocation of key word and intention after the participle, parses key word.Its specific implementation is: to linguistic context, context, user preference and parameters, be intended to combination and weight factor and distribute; The intention key word is carried out giving a mark based on the statistical model of class Markov model, ordering with intention is counter verifies, such as service recorder, custom or other rules by the user, if less than by anti-checking then give certain branch that subtracts; Sort and weight allocation according to intention, determine the intent model that ranks the first.
Structured combinations module 303 is used for the key word that parses is carried out structured combinations, obtains analysis result.
This message processing module 40 comprises:
Authentication module 401 is used for judging whether the domain information of this semantic meaning analysis module is comprehensive, if not comprehensive, sets up this context model, preserves this session log, changes corresponding processing platform over to; If directly change corresponding processing platform over to comprehensively.
" domain information " is the function classification of question answering system definition, and each big function classification all belongs to an independent domain, and a complete domain comprises all information of corresponding function.Need following essential information as: flight domain: the date, (form was: yyyy-MM-dd), set out city or airport, purpose city or airport; In addition, some optional informations, as the classification of attending a banquet in addition, information of discount, airline's information, time interval section (00:00-24:00) etc.
" domain information " checking mainly is the checking of this domain being carried out essential information, see whether essential information all possesses on request, if all possess on request, then verifying that the domain information of passing through directly changes corresponding processing platform over to and carries out the data processing.If do not possess, then check which information of falling vacant, to add Context identifier and corresponding informance field item is added sign according to the domain information of the item of information that lacks, unification is given and is handled and output module.
Processing platform 402 is used for the domain information of checking is handled, and generates the baseline results data.
Join shown in Figure 2ly, Fig. 2 is method flow diagram of the present invention.
In the present embodiment, a kind of answering method based on man-machine combination may further comprise the steps:
Step 1, input user speech, text and operational order are putd question to information, gather customer parameter information;
Step 2, puing question to information and customer parameter information translation with the user is the normative text format information;
Step 3 to text formatting information, is carried out denoising and modular structure processing.
Step 4 is carried out semanteme to user's information of puing question to of denoising and modularization processing and is resolved;
Step 5 is judged and is determined the information processing result;
Step 6 generates the user and puts question to answer, if answer is correct, and directly output; If answer does not have answer or answer is incorrect, submit to artificial customer service to handle problem, judge again and determines according to result, and to judging and deciding ability is trained, export correct option.
Above-mentioned steps 4 comprises following substep:
1, adopt the branch word algorithm that participle is carried out in natural language and the instruction of user's input;
2, to user's linguistic context, context, preference and parameters, carry out first combination and the weight allocation of key word and intention, parse key word;
3, the key word that parses is carried out structured combinations, obtain analysis result.
The invention provides a kind of question answering system based on man-machine combination and method, it is by introducing artifact, adopt artificial instruction mode, self-teaching and the self promotion ability of question answering system have been improved, hommization and intellectuality have been increased, guaranteed accuracy and the promptness of system, made question answering process have more interest and practicality simultaneously, compared pure manual system and saved a large amount of manpower and materials costs.
To those skilled in the art, obviously the invention is not restricted to the details of above-mentioned one exemplary embodiment, and under the situation that does not deviate from spirit of the present invention or essential characteristic, can realize the present invention with other concrete form.Therefore, no matter from which point, all should regard embodiment as exemplary, and be nonrestrictive, scope of the present invention is limited by claims rather than above-mentioned explanation, therefore is intended to include in the present invention dropping on the implication that is equal to important document of claim and all changes in the scope.Any Reference numeral in the claim should be considered as limit related claim.
In addition, be to be understood that, though this instructions is described according to embodiment, but be not that each embodiment only comprises an independently technical scheme, this narrating mode of instructions only is for clarity sake, those skilled in the art should make instructions as a whole, and the technical scheme among each embodiment also can form other embodiments that it will be appreciated by those skilled in the art that through appropriate combination.