|
Description:
|
Readers often encounter the need to explore a document only for a specific point
of interest . We call the phenomena of approaching a narrative not for its entirety , but for
a thread of a particular topic , thematic reading . Present reading tools and information
retrieval techniques provide only limited assistance to readers in such a situation . Our
research centers on this phenomenon . We conducted investigations on both human
behavior and machine automation , with a goal of better meeting the requirements of
thematic reading .
To observe readers ? behavior and understand their expectations , we implemented
a reader ?s interface with designs targeting the predicted needs of thematic readers . We
conducted user studies using both the system and Microsoft Word . We proved that
thematic reading is capable of achieving the goal of understanding a specific topic , at
least to a degree that succeeds in topic -wise tasks . We also reached guidelines for
designing future reading platforms in major aspects such as view , navigation , and
contextual awareness . As for machine automation , we investigated the potential to automatically locate
thematically relevant excerpts . This investigation was inspired by the editorial
compilation of a textbook index . To increase the search performance , we proposed a
two -step methodology which first expands the query with expansion and then filters the
intermediate results by checking the term -occurrence proximity . For query expansion ,
we compared the query expansion with WordNet , morphological inflections , and both
processes together . Our results show that in the context of our study , WordNet made
almost no contribution to the enhancement of recall , while expansion with the
inflectional variants turned out to be a successful and essential scheme . For the
refinement section , the results show that the proximity check on the alternative phrases
formed after inflectional expansion can effectively increase the precision of the
previously acquired return results .
We further tested a different scheme ? using sliding window ? of defining target
and verification units in the methodology . Our findings show that the structural
delimitations (sentences and chapters ) outperformed sliding windows . The first scheme
was able to achieve consistently desirable results , while the results from the second were
inconclusive . |