Conclusion

• Query weighting

  • Dependent on length of query
    • Shorter queries more effective with normalized term freq.
    • Longer queries more effective with tf
  • idf is effective, about equally so as probabilistic factor

• Document weighting

  • Dependent on nature of subject
    • If subject contains more technical terms, enhanced frequency weights are preferable
    • tf more effective for varied vocabulary
    • Binary weighting (0 or 1) more effective for controlled vocabulary
  • idf is most effective
  • Normalization is effective when deviation in number of descriptors is large (as it usually is)