Class Summary for 9/18 from Group #5:

Lauren Barton
Nelson Kile
Caroline O'Hare
Martin Falck
Robert Ryan

The first part of class we reviewed Part B of String Searching Section. In this
section we discussed various methods of string matching, such as:

Naive Algorithm: search is sequential (and bit by bit). The largest number
of comparisons can equal m x n where m = length of string s1 and n= length
of string s2. Used when string is less than 3 chars.

Knuth-Morris-Pratt: One pointer advances in the text the other pointer
advances in the pattern being searched for. Used when alphabet is large.

Boyer-Moore: takes advantage of matches within the pattern being searched for.

Boyer-Moore-Horspool: Is used most often

Shift-Or: used for comparing regular expressions

Karp-Rabin: Uses hashing for string matching


The second part of the lecture reviewed the first three sections of the RR Unit:
We discussed the basic concern of improve the retrieval process to bring
back a higher success rate in precision (or the retrieval of relevant
documents). Ranking is a process used to retrieve the documents with a
higher probability of relevance. Certain features can be used to rank
documents, such as term weighting.
One model in particular was examined which is used to rank documents. This
is the Vector Space Model. In this model documents are ranked based on their
similiarity to each other. Feedback is used to tune query and provide even
more relevant feedback.
A second model examined was the Probabilistic Model. This model is based on
the assumption that terms should be given greater weights if they appeared
in documents retrieved from a previous query. So if a user is attempting to
locate documents on a particular topic it makes sense to assume that the
documents retrieved in an earlier query are related to the present query.

= = = = = ==  =  = === =  = =  = = = = =   =  ====  = = == =  = =  == = =
Tom Kalafut
09/18/95 class summary

    We glanced over several algorithms for string searching with emphasis
on why or when you would each type of algorithm.  We then started some
relevance ranking material including sub-vectors, query moving, and
probabilistic algorithms.