Personalized information delivery: an analysis of information filtering methods

10345 Abstract 10346 Abstract 10347 Abstract 10348 :iltered Abstracts Abstract 10349 Abstract 1040210349 Abstract 10402 Highly rated abstracts added to document profile each month T h e score for e ach new T M was t he cos ine b e t w e e n t he T M vec tor a n d the nearest i n t e r e s t vector . T h e new T M s were t h e n r a n k e d based o n the i r m a x i m u m cosine score. T h u s , T M s o c c u r r i n g close to any p o i n t o f i n t e r e s t fo r a p a r t i c u l a r e m p l o y e e would be r a n k e d t he h ighes t . T h e same type o f c o m p a r i s o n was d o n e fo r the d o c u m e n t prof i les . Each docu m e n t in t he d o c u m e n t p rof i l e was r e p r e s e n t e d as a s epa ra t e vec tor a n d c o m p a r e d to all new T M s . T h e new T M s were t h e n r a n k e d based o n the m a x i m u m cosine for e ach T M to any d o c u m e n t in t he d o c u m e n t prof i le . T h e s e c o m p a r i s o n s r e su l t ed in f o u r r a n k o r d e r e d lists o f abstracts , o n e for e ach f i l t e r ing m e t h o d . T h e top abs t rac t s f r o m each m e t h o d were t h e n sen t to each emp loyee , w h o r a t e d t h e m fo r r e l evance to his o r h e r t echn ica l in teres ts . Fo r each m e t h o d , the t op 25% o f t he abs t rac t s r a t e d 4 or h i g h e r were t h e n i n c o r p o r a t e d in to t he e m p l o y e e s ' d o c u m e n t p ro files. 3 I n the f i rs t m o n t h o f s tudy, on ly the w o r d prof i l e was u sed s ince e m p l o y e e s h a d no t p rev ious ly indica ted which abs t rac t s to i nc lude in t h e i r d o c u m e n t prof i le . I n the subseFlOure 11. The f i l t e r i n g p r o c e s s Table 1. The four f i l ter ing methods