A kind of deep learning intelligent response system of modified based on computer cloud data
Technical field
The present invention relates to intelligent answer robotics, more particularly to a kind of modified is based on computer cloud data
Deep learning intelligent response system.
Background technology
Based on the mobile Internet platform of current popular, the research for carrying out intelligent answer robot system is one emerging
Research field.The research field being related to includes based on mobile Internet platform, user in campus real time position it is dynamic
The Changing Pattern of state variation monitoring, User Activity position and scope, the regular course of User Activity is defined and pushed away with personalized task
Recommend, the fusion of the artificial intelligence knowledge base in combination with user context key element and question and answer expert system etc. all with mobile Internet
On the related technical field of campus intelligent answer robot.
Not yet there is the campus intelligent answer machine in combination with big data excavation for mobile Internet in prior art
The correlation technique of device people's system.
The content of the invention
In view of this, the present invention proposes a kind of deep learning intelligent response system of modified based on computer cloud data.
A kind of deep learning intelligent response system of modified based on computer cloud data, it is included such as lower unit:
Positioning unit, for positioning to user's real time position, is sent to real-time position information cartographic information and describes
Unit;
Cartographic information delineation unit, for receiving the real-time position information of positioning unit transmission, and real-time position information shape
Into the action trail of user, the action trail of user is segmented, and trajectory segment is described;Based in Preset Time
The description information of the trajectory segment of user delimit User Activity region;It is additionally operable to depict the complete map of specific region in advance,
And on map detailed each point of interest marked in the specific region, be a certain specific in the real-time position information of user
During point of interest, the relevant information of the particular point of interest is pushed;
Information receiving unit, for the text query or speech query information of receiving user's input, is believing for text query
During breath, through the text mining and process of Chinese word segmentation, semantic analysis and syntactic analysis, and the information after process is sent to into number
According to library unit;For speech query information when, first speech query information is identified as into text query information, and through Chinese point
The text mining and process of word, semantic analysis and syntactic analysis, and the information after process is sent to into Database Unit;For text
During this Query Information, the information after process is sent to into Emotion identification unit, for speech query information when, while will process after
Information and speech query information be sent to Emotion identification unit;
Emotion identification unit, the information for being sent according to information receiving unit judges the emotional state of user, user's
Emotional state is divided into general, low, sad, happiness;For it is low or sad when, the relevant information of particular point of interest is pushed strong
The instruction that degree is turned down is sent to cartographic information delineation unit;
The Emotion identification unit includes following word Emotion identification subelement and voice mood identification subelement:
Word Emotion identification subelement includes that classification storage represents general, low, sad, glad standard vocabulary;
By classifier functions the vocabulary of test input is trained and is categorized in each classification, and according to classification
As a result judgment models are set up;
The information of reception is matched with each single linguistic context of target word in standard vocabulary, semantic contrast is caught
Information, and the mood of Word message is judged to acquired semantic comparative information;
Voice mood identification subelement identification subelement is used to recognize the emotional information of voice by speech recognition technology;
Database Unit, proposes that problem and the answer knowledge material of answer set up knowledge for collecting and arranging each user
Storehouse, to story extraction candidate keywords, and the content to keyword carries out handmarking, to carry out follow-up keyword rope
Draw;It is problem knowledge storehouse and answer knowledge base by Knowledge base partition, is used to deposit each of user's proposition wherein in problem knowledge storehouse
The demand of kind and problem, there are problems that and the demand in problem knowledge storehouse and the answer content for matching, answer in answer knowledge base
Thinking and answer detailed process;And the association that question and answer matching degree is assessed is set up between problem knowledge storehouse and answer knowledge base, close
The matching degree being combined between metric question and answer, and by this process in combination with follow-up user satisfaction is fed back,
It is provided commonly for the optimization to knowledge base, and the renewal to related content in problem knowledge storehouse and answer knowledge base and optimization;
Social networks describes unit, for obtaining interrelated between the user in specific region and influencing each other and use
The different social role of performer between family, so that it is determined that each user each has different user property and user task,
And the interaction between user constitutes respective customer relationship and social circle.
Modified of the present invention based on computer cloud data deep learning intelligent response system in, it include as
Lower unit:
Statistical analysis unit, for according to the description information of the trajectory segment of user, each user in Preset Time each
Respective customer relationship and social circle's life are constituted with the interaction between different user property and user task, and user
Into and preserve the stroke rule information table of user;Include user within each time period in the stroke rule information table of user
The interest of zone of action, activity item and user.
In deep learning intelligent response system of the modified of the present invention based on computer cloud data, the map
Information portrayal unit is additionally operable to when the real-time position information of positioning unit transmission meets the point of interest in specific region, and this is emerging
Interest point is sent to statistical analysis unit;
Statistical analysis unit is judged in the time point in point of interest item whether to be handled;There is backlog
When, sent to information receiving unit with the interest point name and backlog;
Information receiving unit through Chinese word segmentation, semantic analysis and syntactic analysis text mining and process, and will process
Information afterwards is sent to Database Unit;
The answer content matched with problem, answer thinking and answer detailed process are pushed and shown by Database Unit.
In deep learning intelligent response system of the modified of the present invention based on computer cloud data,
The processing authority of the backlog of user is prestored in social networks describes unit, processing authority is divided into certainly
Oneself processes and allows specific user's alternate process in social circle.
In deep learning intelligent response system of the modified of the present invention based on computer cloud data,
Statistical analysis unit there is backlog in certain time period for user and user is in backlog
Zone of action outside when, judge that whether the backlog is the pending thing for allowing specific user's alternate process in social circle
, if it is the backlog is sent into user social contact circle corresponding other users.
In deep learning intelligent response system of the modified of the present invention based on computer cloud data,
Matching degree between Database Unit vacuum metrics question and answer, and this process is satisfied with follow-up user
Degree feedback combines, and is provided commonly for the optimization to knowledge base, and to related content in problem knowledge storehouse and answer knowledge base more
Newly include with optimization:
Matching mapping relations will be set up between question and answer after question and answer matching degree evaluation process, and to candidate answers
Ranking is carried out with the matching degree of problem, is supplied to user to be selected several candidate answers in the top, while allowing
Answer of the user to problem is evaluated, and the result of evaluation is fed back in time into question answering system carries out the amendment of question and answer matching degree
Replacement with knowledge in knowledge base and renewal.
In deep learning intelligent response system of the modified of the present invention based on computer cloud data,
Mapping relations matrix M in Database Unit between problem knowledge storehouse and key to the issue wordq,kIt is expressed as follows:
Wherein Rq→kRepresent mapping relations therebetween, Qi(i=1...n) problem of representation and its number, Kj(j=
1...m) problem of representation keyword and its number;One problem can be disassembled and is been described by for multiple keywords, while one
Keyword can also apply in multiple problems;Tij(i=1...n, j=1...m) problem of representation i is by key to the issue word j institutes group
Into content of text;
Mapping relations matrix M in Database Unit between key to the issue word and answer keywordk,k’It is expressed as follows:
Wherein Rk→k'Represent mapping relations therebetween, Ki(i=1...n) problem of representation keyword and its number, K'j
(j=1...m) answer keyword and its number are represented;One key to the issue word can correspond to multiple problem answers keywords,
Simultaneously an answer keyword can also apply in multiple key to the issue words;Tij(i=1...n, j=1...m) problem of representation
The content of text that keyword i is answered by answer keyword j;
Mapping relations matrix M in Database Unit between answer keyword and answerk’,aIt is expressed as follows:
Wherein Rk'→aRepresent mapping relations therebetween, K'i(i=1...n) answer keyword and its number, A are representedj
(j=1...m) answer and its number are represented;One answer keyword can be to describe multiple answers, while an answer
Can be described with multiple answer keywords;Tij(i=1...n, j=1...m) represents that answer keyword i enters to answer j
The content of text of row description;
Mapping relations matrix M in Database Unit between answer and answer rankinga,rankIt is expressed as follows:
Wherein Ra→rankRepresent mapping relations therebetween, Ai(i=1...n) answer and its number are represented,Represent the ranking of answer i;
Mapping relations matrix M in Database Unit between answer and user satisfactiona,usIt is expressed as follows:
WhereinRepresent mapping relations therebetween, Ai(i=1...n) answer and its number, U are representeds,j(j=
1...m) user satisfaction and its number are represented;One answer can be evaluated by multiple users, can possess multiple users
Satisfaction, while a user can also be evaluated multiple answers, provides respectively the customer satisfaction evaluation of multiple answers;
Pij(i=1...n, j=1...m) represents the user satisfaction that answer i is evaluated by user j.
Deep learning intelligent response system and existing skill of the modified of present invention offer based on computer cloud data is provided
Art is compared and had the advantages that:For aiding in specific region, the new user for adding (such as teaches (such as in campus)
Teacher and student etc.) faster be familiar with and adapt to environment (University Environments and cultural atmosphere of such as user place universities and colleges), preferably with
Friend relation is set up between other users, and the artificial intelligence knowledge base set up by situation key element is movable crowd in specific region
Daily life preferably service.So that user faster, is more preferably familiar with specific region internal and external environment, hardware/software infrastructure, area
Domain culture etc., and systematicness is carried out with question and answer mode represent with a series of.Can be so that according to the mood of user, it is strong that adjustment is pushed
Degree, hommization degree is higher.
Description of the drawings
Fig. 1 is the deep learning intelligent response system architecture frame of the modified based on computer cloud data of the embodiment of the present invention
Figure.
Specific embodiment
As shown in figure 1, for the defect of prior art, the present invention proposes a kind of modified based on computer cloud data
Deep learning intelligent response system, it is included such as lower unit:
Positioning unit, for positioning to user's real time position, is sent to real-time position information cartographic information and describes
Unit.Alternatively, positioning unit can pass through the function that the app such as mobile phone realize positioning.
Cartographic information delineation unit, for receiving the real-time position information of positioning unit transmission, and real-time position information shape
Into the action trail of user, the action trail of user is segmented, and trajectory segment is described;Based in Preset Time
The description information of the trajectory segment of user delimit User Activity region;It is additionally operable to depict the complete map of specific region in advance,
And on map detailed each point of interest marked in the specific region, be a certain specific in the real-time position information of user
During point of interest, the relevant information of the particular point of interest is pushed.It is alternatively possible to the algorithm for being based on user collaborative filtration pushes the spy
Determine the relevant information of point of interest.Such as through some point of interest, when being coffee, then coffee phase in annex preset range is pushed
The information of pass is to user.
The action trail of user is segmented, and trajectory segment is described, for example, user is if a university
It is raw, the trajectory segment from dormitory to teaching building is described as " turning out for work ".These are only schematically.
Information receiving unit, for the text query or speech query information of receiving user's input, is believing for text query
During breath, through the text mining and process of Chinese word segmentation, semantic analysis and syntactic analysis, and the information after process is sent to into number
According to library unit;For speech query information when, first speech query information is identified as into text query information, and through Chinese point
The text mining and process of word, semantic analysis and syntactic analysis, and the information after process is sent to into Database Unit;For text
During this Query Information, the information after process is sent to into Emotion identification unit, for speech query information when, while will process after
Information and speech query information be sent to Emotion identification unit;
Emotion identification unit, the information for being sent according to information receiving unit judges the emotional state of user, user's
Emotional state is divided into general, low, sad, happiness;For it is low or sad when, the relevant information of particular point of interest is pushed strong
The instruction that degree is turned down is sent to cartographic information delineation unit;
The Emotion identification unit includes following word Emotion identification subelement and voice mood identification subelement:
Word Emotion identification subelement includes that classification storage represents general, low, sad, glad standard vocabulary;
By classifier functions the vocabulary of test input is trained and is categorized in each classification, and according to classification
As a result judgment models are set up;
The information of reception is matched with each single linguistic context of target word in standard vocabulary, semantic contrast is caught
Information, and the mood of Word message is judged to acquired semantic comparative information;
Voice mood identification subelement identification subelement is used to recognize the emotional information of voice by speech recognition technology.
Alternatively, the formula of judgment models is as follows:
Wherein, w is target classification word in context window, and c is the input vocabulary in situational meaning;K is existing for negative word remittance abroad
Probability;Sim represents similarity cosine value, and σ is Dynamic gene.
Database Unit, proposes that problem and the answer knowledge material of answer set up knowledge for collecting and arranging each user
Storehouse, to story extraction candidate keywords, and the content to keyword carries out handmarking, to carry out follow-up keyword rope
Draw;It is problem knowledge storehouse and answer knowledge base by Knowledge base partition, is used to deposit each of user's proposition wherein in problem knowledge storehouse
The demand of kind and problem, there are problems that and the demand in problem knowledge storehouse and the answer content for matching, answer in answer knowledge base
Thinking and answer detailed process;And the association that question and answer matching degree is assessed is set up between problem knowledge storehouse and answer knowledge base, close
The matching degree being combined between metric question and answer, and by this process in combination with follow-up user satisfaction is fed back,
It is provided commonly for the optimization to knowledge base, and the renewal to related content in problem knowledge storehouse and answer knowledge base and optimization.
Social networks describes unit, for obtaining interrelated between the user in specific region and influencing each other and use
The different social role of performer between family, so that it is determined that each user each has different user property and user task,
And the interaction between user constitutes respective customer relationship and social circle.Alternatively, customer relationship and social circle can draw
Graduation, the user in different grades of social circle has different information browse authorities.
Modified of the present invention based on computer cloud data deep learning intelligent response system in, it include as
Lower unit:
Statistical analysis unit, for according to the description information of the trajectory segment of user, each user in Preset Time each
Respective customer relationship and social circle's life are constituted with the interaction between different user property and user task, and user
Into and preserve the stroke rule information table of user;Include user within each time period in the stroke rule information table of user
The interest of zone of action, activity item and user.
In deep learning intelligent response system of the modified of the present invention based on computer cloud data, the map
Information portrayal unit is additionally operable to when the real-time position information of positioning unit transmission meets the point of interest in specific region, and this is emerging
Interest point is sent to statistical analysis unit.
Statistical analysis unit is judged in the time point in point of interest item whether to be handled;There is backlog
When, sent to information receiving unit with the interest point name and backlog.
Information receiving unit through Chinese word segmentation, semantic analysis and syntactic analysis text mining and process, and will process
Information afterwards is sent to Database Unit.
The answer content matched with problem, answer thinking and answer detailed process are pushed and shown by Database Unit.
In deep learning intelligent response system of the modified of the present invention based on computer cloud data,
The processing authority of the backlog of user is prestored in social networks describes unit, processing authority is divided into certainly
Oneself processes and allows specific user's alternate process in social circle.Implement the present embodiment, the use of user can be greatly improved just
Profit.
In deep learning intelligent response system of the modified of the present invention based on computer cloud data,
Statistical analysis unit there is backlog in certain time period for user and user is in backlog
Zone of action outside when, judge that whether the backlog is the pending thing for allowing specific user's alternate process in social circle
, if it is the backlog is sent into user social contact circle corresponding other users.Embodiment the present embodiment, is carrying
While high convenience for users, security is also improved.
In deep learning intelligent response system of the modified of the present invention based on computer cloud data,
Matching degree between Database Unit vacuum metrics question and answer, and this process is satisfied with follow-up user
Degree feedback combines, and is provided commonly for the optimization to knowledge base, and to related content in problem knowledge storehouse and answer knowledge base more
Newly include with optimization:
Matching mapping relations will be set up between question and answer after question and answer matching degree evaluation process, and to candidate answers
Ranking is carried out with the matching degree of problem, is supplied to user to be selected several candidate answers in the top, while allowing
Answer of the user to problem is evaluated, and the result of evaluation is fed back in time into question answering system carries out the amendment of question and answer matching degree
Replacement with knowledge in knowledge base and renewal.
In deep learning intelligent response system of the modified of the present invention based on computer cloud data,
Mapping relations matrix M in Database Unit between problem knowledge storehouse and key to the issue wordq,kIt is expressed as follows:
Wherein Rq→kRepresent mapping relations therebetween, Qi(i=1...n) problem of representation and its number, Kj(j=
1...m) problem of representation keyword and its number;One problem can be disassembled and is been described by for multiple keywords, while one
Keyword can also apply in multiple problems;Tij(i=1...n, j=1...m) problem of representation i is by key to the issue word j institutes group
Into content of text;
Mapping relations matrix M in Database Unit between key to the issue word and answer keywordk,k’It is expressed as follows:
Wherein Rk→k'Represent mapping relations therebetween, Ki(i=1...n) problem of representation keyword and its number, K'j
(j=1...m) answer keyword and its number are represented;One key to the issue word can correspond to multiple problem answers keywords,
Simultaneously an answer keyword can also apply in multiple key to the issue words;Tij(i=1...n, j=1...m) problem of representation
The content of text that keyword i is answered by answer keyword j;
Mapping relations matrix M in Database Unit between answer keyword and answerk’,aIt is expressed as follows:
Wherein Rk'→aRepresent mapping relations therebetween, K'i(i=1...n) answer keyword and its number, A are representedj
(j=1...m) answer and its number are represented;One answer keyword can be to describe multiple answers, while an answer
Can be described with multiple answer keywords;Tij(i=1...n, j=1...m) represents that answer keyword i enters to answer j
The content of text of row description;
Mapping relations matrix M in Database Unit between answer and answer rankinga,rankIt is expressed as follows:
Wherein Ra→rankRepresent mapping relations therebetween, Ai(i=1...n) answer and its number are represented,Represent the ranking of answer i;
Mapping relations matrix M in Database Unit between answer and user satisfactiona,usIt is expressed as follows:
WhereinRepresent mapping relations therebetween, Ai(i=1...n) answer and its number, U are representeds,j(j=
1...m) user satisfaction and its number are represented;One answer can be evaluated by multiple users, can possess multiple users
Satisfaction, while a user can also be evaluated multiple answers, provides respectively the customer satisfaction evaluation of multiple answers;
Pij(i=1...n, j=1...m) represents the user satisfaction that answer i is evaluated by user j.
Deep learning intelligent response system and existing skill of the modified of present invention offer based on computer cloud data is provided
Art is compared and had the advantages that:For aiding in specific region, the new user for adding (such as teaches (such as in campus)
Teacher and student etc.) faster be familiar with and adapt to environment (University Environments and cultural atmosphere of such as user place universities and colleges), preferably with
Friend relation is set up between other users, and the artificial intelligence knowledge base set up by situation key element is movable crowd in specific region
Daily life preferably service.So that user faster, is more preferably familiar with specific region internal and external environment, hardware/software infrastructure, area
Domain culture etc., and systematicness is carried out with question and answer mode represent with a series of.
It is understood that for the person of ordinary skill of the art, can be done with technology according to the present invention design
Go out other various corresponding changes and deformation, and all these changes and deformation should all belong to the protection model of the claims in the present invention
Enclose.