Grading criteria for Midexam

1)

    a)

        Precision = 0.5     1.5 points

        Recall = 0.4          1.5 points

        MAP = 0.38          2 points

    b)

     i) Writing down the “MaxDCG” formula:   2 points

    ii) Reasonable explanation why do we need this normalizer.  2 points

    iii) We do normalization in MAP.  1 point

2)

    a) TF, IDF, Doc normalization, each 1.5 points

    b) IDF: 1 point    correct explanation: 2 points

    c) 2 points for each role

    d) Hub and Authority 2 points for each

    e) 3.5 points for correct explanation

3)

    a)  1 point for each item

    b)  3 points for mentioning H(Xa) and H(Xc) and 2 points for correct explanation

    c)  3 points for correct mentioning and 2 points for explanation

    d)  4 points for writing down Bayes formula, 2 points for decomposition, 2 points for correct conclusion

4)

    a) one point for each correct probability

    b) 2.5 points for each

    c) 2 point for using the correct probability p(w4|d) and p(w6|d), 5 points for multiplication instead of summation

    d) 4 points for each correct query, 2 points deducted if you mentioned vice versa

5)

    a) 1 point deducted for minor mistake

    b) 5 points for correct explanation, zero if said “NO”

6)   Items     1  point

     Other users          1 point

     Rating   1 point

    Algorithm:  7 points:

     Showing knowledge about how memory-based method for collaborative filtering works 2 points
                 Proposing a reasonable algorithm to improve retrieval accuracy: 3 points;
                Giving fairly detailed formulas: 2 points

7)  Attempted with some idea(s): 1
                Optimize the utility or the decision: 2
                Giving a solution for theta: 2 (regardless correctness)
                Giving a correct solution: 3
                Showing derivation of the solution: 2