id: 01936196
dt: j
an: 01936196
au: Lee, Dik Lun; Ren, Liming
ti: Document ranking on weight-partitioned signature files.
so: ACM Trans. Inf. Syst. 14, No. 2, 109-137 (1996).
py: 1996
pu: Association for Computing Machinery (ACM), New York, NY
la: EN
cc: H.3.3 H.2.2 H.3.6 H.3.1
ut: algorithms; design; experimentation; performance
ci: 
li: doi:10.1145/226163.226164
ab: Summary: \BeginparA signature file organization, called the
    weight-partitioned signature file, for supporting document ranking is
    proposed. It employs multiple signature files, each of which
    corresponds to one term frequency, to represent terms with different
    term frequencies. Words with the same term frequency in a document are
    grouped together and hashed into the signature file corresponding to
    that term frequency. This eliminates the need to record the term
    frequency explicitly for each word. We investigate the effect of false
    drops on retrieval effectiveness if they are not eliminated in the
    search process. We have shown that false drops introduce insignificant
    degradation on precision and recall when the false-drop probability is
    below a certain threshold. This is an important result since false-drop
    elimination could become the bottleneck in systems using fast signature
    file search techniques. We perform an analytical study on the
    performance of the weight-partitioned signature file under different
    search strategies and configurations. An optimal formula is obtained to
    determine for a fixed total storage overhead the storage to be
    allocated to each partition in order to minimize the effect of false
    drops on document ranks. Experiments were performed using a document
    collection to support the analytical results.% \Endpar (Provider: ACM)
    Review: \BeginparA modified vector-space technique for ranking
    documents during retrieval is described. The ranking is based on
    similarity to a user-supplied query string. In order to improve
    retrieval response time on large repositories, the authors introduce
    weighted-partitioned signature files, with one signature file per term
    frequency. With an optimal signature configuration, postprocessing
    ({“}false-drop{”} elimination) can be removed by assigning
    different lengths to the signature files. The lengths are dependent on
    the frequency weights of the files. Higher-frequency files have longer
    lengths. Therefore, the authors show that, by eliminating term
    postprocessing, they can still reduce the effect of false-drops without
    sacrificing recall and precision effectiveness.\Endpar \BeginparWhile
    the authors briefly describe both inversion and signature files, they
    do not discuss existing vector space and clustering models [1]. Their
    approach is a modification to the existing vector space techniques.
    They introduce an implied search order of vector groups produced by a
    document. Each signature grouping has a common term frequency and
    signature length. As a result, any document-to-document vector
    similarity function, such as the cosine similarity function, can be
    used. A comparison to the existing vector cluster searching methods
    would have been appropriate. One example is latent semantic indexing
    (LSI), which applies singular value decomposition [2].\Endpar
    \BeginparOverall, the paper{’}s ideas are well organized and well
    presented. I recommend it to anyone interested in document retrieval
    techniques. The authors{’} idea of weighted-partitioned files is a
    good extension of vector models that support weight-based querying.
    However, the lack of comparison to other indexing techniques, such as
    LSI, which can use spatial access methods, is a significant
    flaw.\Endpar (Provider: ACM)
rv: