Model Personalization 1 : Data Fusion

1
Model Personalization (1) Data Fusion

• Improve frame and answer (of persistent query)
  generation through Data Fusion (local fusion on
  personal and topical level) and Interactive
  Relevancy Feedback.
• At stage 1, we have successfully integrated
  effective data fusion into HITIQA to optimize the
  successful rate of useful paragraph extraction.
  At stage 2, the emphasis will be on using user
  judgments at different times to adjust fusion
  parameters chronologically, with a time-sensitive
  weighting scheme, to fit the evolving
  understanding of the same user on the topic.

2
Model Personalization (2) Document Qualities
Judgment

• Personalization of automatic document qualities
  assessment algorithm, through advanced
  statistical analysis and machine learning, to
  identify (1) global qualities predictors, (2)
  general formal model of qualities assessment, and
  (3) personal weight on parameters for individual
  preference.
• At stage1, we have established a few models in
  estimation of various document qualities, based
  on textual features and linguistic patterns of a
  document, with successful rate much better than
  chance, on a global level. At stage 2, we will
  continue our previous endeavor in identification
  of good predictive variables of qualities, with a
  new emphasis on a local level to mimic the
  personal mental model of a user.

3
Model Personalization (3) Integration through
Experiment

• We will integrate the previous two
  personalization and other desired mechanisms
  into a single interface, by converting related
  functionalities into position and iconic
  information in the user display.
• At stage 2, focusing on same user and persistent
  query, we will investigate the impacts of
  Interface Options on analyst satisfaction and
  identify the best combination strategy, and to
  establish the effectiveness measure on a personal
  level.
• In addition to ANOVA and multiple comparisons
  such as Tukeys method, we will use orthogonal
  arrays method to reduce the number of
  experimental configuration to be studied. Instead
  of trying to identify the causes of negative
  effect, we will focus on how to neutralizes
  negative effect to obtain a higher quality result
  with fewer experiments.