Daily summary 9/19
Chris Ye
        Today we discussed the unit SS part B in class. Mainly, we
concentrated on "Shift-Or" operation. The "Shift-Or" operation represents
the position of a entry which occurs in the pattern.
        Also we started the Unit RR with Basic Vector Space Model. The
example in the note is three dimension vector space, however we can
easily scale up to n dimension vector space. In our three dimension case,
we have system, information, and retrieval at each dimension. A
document is located in somewhere of this three dimension space. The
location is depended on the relative relationship between each axis.
Now we are able to tell how a document is related to others by looking
those positions. Also we are able to position a query in vector space as
same as the documents
        To measure the similarity of two documents, we borrowed the
concept of vector product. The similarity between two points is measured
by the cosine of the angle between their two vectors.
        At the end of class, we went through the probability model.

= = = = = ==  =  = === =  = =  = = = = =   =  ====  = = == =  = =  == = =
Class Summary, Sep 19 -- Srinivas R. Gaddam

One of the String Matching algorithms, the Shift-Or algorithm, was
explained. The difference between this algorithm and others are that here
the pattern is preprocessed and not the text. This algorithm has some
advantages over the other algorithms and is suitable for hardware
implementation.
 
The unit 5 was then continued.
There are two models to make use of the different features -- Vector Space
Model and Probabilistic Model.
Vector Space Model: In this model, each document is represented by a point
in a t-dimensional vector space, where each dimension represents a term or
concept found in documents. A set of documents can be represented by a
Centroid point. In this model, a query can also be represented by a point and
as such can be treated as a document. This increases the model's functionality
as it facilitates interchange of queries and documents which the boolean model
doesn't allow. This way an entire document can be used as query.
When a query is given, it is located in the vector space and the set of all
the documents "near" this point is the result of the query.
Similarity between two points is measured by the cosine of the angle between
the query and document vectors.
Vector Feedback can be used to improve the results obtained. There are three
scenarios:
retrospective - everything is known including relevance
predictive -  results from initial searches, some relevant documents are used
formulas :Rocchio - this method uses the info about all the nonrelevant
documents, whereas Ide method uses only the nonrelevant documents with the
highest value.
One of these methods is used to shift the point that represents the query near
the documents that are relevant. Normally, three such shifts (or iterations)
are sufficient to get the result.

Probabilistic Model: The assumption that should be made about the information
retrieval theory to apply the probability theory is mentioned and justified.
It is "The relevance of a document to a request is independant of the other
documents in the collection." This is justified because there are so many
documents that, their dependance on one another can be ignored.
Probability ranking principle states that if the documents resulting from
a query are ranked in the decreasing probability of relevance to the user,
then it is the best possible solution the query.