我写了那么多report

We are going to discuss the background information we used. The idea of matching words from
multiple documents is inherently complex due to the sophistication of human language. For example
in the English dictionary, there are many words that map to multiple definitions. This provides a
challenge when attempting to match text and meaning. One way of dealing with this is proposed
by Berry et al, which is using Latent Semantic Indexing (LSI) [2]. LSI is an information retrieval
method that uses Singular Value Decomposition (SVD) to identify patterns between terms and
concepts in a set of texts. So in order to implement LSI, a matrix must be construct whose
elements are each of the terms in the various documents. For example denote
A = [aij ]
where aij denotes the frequency in which term i occurs in document j. This means that A is likely
to be a very sparse matrix because most documents do not contain every word we are searching
for. After creating a term by document matrix,we can compute the SVD of this matrix, create
the database of singular values and vectors for retrieval, and match user queries to documents.
In addition, constructing a dictionary vector which containing all the words in documents are
helpful. The singular value decomposition (SVD) will estimate the structure in word usage across
documents.
We also need to create a m 1 query vector q. Inside the vector q, the position of the number
1 in an entry represents the term occurs once in the query vector. For purposes of information
retrieval, a user’s query can be represented as a vector q in k-dimensional or n-dimensional space.
A query (like a document) is a set of words. The length of the query vector q will be k or n.
The number in the query vector represents the occurrence of a specific word in the dictionary. For
example, the user query can be represented by
q = q
TUk
1
k
The query vector will used to compared with all existing document vectors in the matrix A and and
find most relevant documents.For example, the most relevant to query terms could be computed
by y = AT
q, but actually we approach the result in a different way.
Furthermore, we will use measure the cosine between the query vector and document vector
to find the similarity between them since there may be documents that are more relevant than
another. Using the value of cosine as a measure tells us what’s the better match. For example, the