I would like to design an intelligent response system
I designed before the idea is to use <What did you say?> Corresponds to <What I reply>?
Then add the statement data plus full-text index on the fuzzy search
The reply system, but this design can really do correspond to any <What did you say?>
I can back out of a not speak the same sentence.
I feel that the limited capacity of the system replies.
But recent stop to spend time in the database, data mining, in fact, SQL SERVER (I use the SQL SERVER2008R2)
Can be by the application of data mining to do some intelligent.
It's all thanks to the SQL SERVER built-in powerful algorithms in data mining.
Before you spend a lot of time to get them to study the spirit and content of these algorithms.
There are two algorithms I think the feasibility of this intelligent reply system.
In the case of the development of great help to a decision tree algorithm is the neural network algorithm have to find out but before
Analysis Services the main function of 1.Classification
2.Estimate 3.Forecast 4.Associated with packet 5.Homogeneous grouping to do classification.
When I Think while connotation of these two algorithms and the spirit, I am also thinking about how these two algorithms applied to the CASE.
But there are a few basic questions have to understand clearly
First first decision tree algorithms focus on the dominant classification neural network algorithm is to do implicit classification.
This is a great difference of the two algorithms.
So if I use the decision tree algorithm used in this CASE.
I think the direction is: how do the topic classification in accordance with the sentence structure and so on.
But soon, I understand that we humans use sentences doing communication.
To do the work of the classification is very complex, it seems difficult to do the work of the classification is perfect.
So I think that the use of decision trees seems a little difficulty.
Compared with the algorithms in neural network, its approach is a and recessive classification.
That is to observe the model's input and output to determine the classification categories.
This work is in dealing with its background by the algorithm, I do not need to worry about how.
As long as I am concerned about the classification of this work is right or wrong.
So quickly, I realized that I really want to do things.
Seem to get through the neural network algorithm to accomplish.
So I tend to use neural network design algorithm to get the job done.
But I have no experience operating in the Analysis Services Practice.
I was the first to consider this spirit to their design work.
Therefore, I first think in the design of the establishment of three rows <What did you say?> <What I reply> <Classification of right / wrong?>
Store large amounts of <You said what?> With <What I reply> Sentence <classification of right / wrong> decision-making factor for neural network
Use of <What did you say?> Corresponds to <What I reply> Randomly to the corresponding sentence.
And then use the <and the classification of right / wrong> under the corresponding sentence is right or wrong.
Information on the number of decision-making so that when thrown into the data enough large enough the neural network is also a solid enough.
Can automatically sentence intelligent reply system out to do a both intelligent response system.
This is my idea.
I do not know you predecessors have any comment please feel free to enlighten me!
Hi 向恩, to be able to answer your questions, we need to narrow them down:
Do you want to create artificial intelligence voice system? What are the requirements in terms of response time and data volume? Are you planning to use SSAS for this project?
I re-Restatement of my thoughts.
I want to do is intelligent reply system, not by adding the function of speech.
I design ideas I have built a large number of verbal exchanges Reply data in SQL.
Focus impossible fluent reply in accordance with the existence of SQL.
Although I can use full-text index search technology allows SQL to become more flexible in the search for verbal exchanges Reply
My word there are a few keywords, I can provide an appropriate reply to this a few keywords.
But this is not the best solution in this application.
I have to teach to do a proper classification of auxiliary and responding to the logic on the PC of the content of the reply.
So I think that the SSAS is an optimal solution.
Using neural network algorithm.
Therefore, analysis of the data that I provided to the SQL-fluent reply data.
Analysis of verbal exchanges Reply The quality of the information as the eigenvalue of the artificial neural network algorithm.
Used to establish a good network of neurons.
I hope you can give me good advice, thank you.
I am sorry I am having difficulty understanding the detail about your question. Can you please provide some examples of data you want to analyze?
If you preferred to communicate using your native language, please post your question in your native language. I can read and write Chinese, and I know some one who can read Japanese too.
- Изменено Haidong HuangMicrosoft employee 20 февраля 2012 г. 23:44 typo
Hi, Haidong Huang
Thank you for your interest in my question, my question the use of Chinese re-described.
I will throw a my skydriver space.