Query Suggestion @ tokyotextmining#2
•Download as KEY, PDF•
7 likes•1,718 views
see also http://d.hatena.ne.jp/repose/20100927/1285524737 tokyotextmining#2 @y_benjo
Recommended
More Related Content
What's hot
What's hot (18)
Viewers also liked
Viewers also liked (20)
Similar to Query Suggestion @ tokyotextmining#2
Similar to Query Suggestion @ tokyotextmining#2 (20)
More from ybenjo
More from ybenjo (16)
Recently uploaded
Recently uploaded (20)
Query Suggestion @ tokyotextmining#2
- 17. Suggestion Reformulation http://www.london-eating.co.uk/newsletter/2006/may/images/wine.jpg
- 18. Teaching Math or Something / foundphotoslj
- 22. • Mei Q, Zhou D, Church K. Query suggestion using hitting time. Proceeding of the 17th ACM conference on Information and knowledge mining - CIKM '08. 2008:469. • Herdagdelen A, Ciaramita M, Mahler D, et al. Generalized syntactic and semantic models of query reformulation. In: Proceedings of the 33rd ACM SIGIR Conference, Geneva, Switzerland.; 2010.
- 23. Query suggestion using hitting time (CIKM 2008)
- 25. introduction (1)
- 26. introduction (2)
- 28. V1 W V2
- 29. P V1 W V2 w(i, j) pij = di di = w(i, j) j∈V2 30 p11 = 52 15 p12 = 52 7 p13 = 52
- 31. V1 V1 V2 V1 w(i, k) w(k, j) pij = di dk k∈V2 = pik pkj k∈V2
- 33. A V Xt t TA A T A = min{t : Xt ∈ A, t ≥ 0}
- 34. i P [T A = m|X0 = i] = P [X1 = j|X0 = i] j∈V ·P [T A = m − 1|X0 = j] = pij P [T A = m − 1|X0 = j] j∈V TA = m m-1 A
- 35. A X0 = i TA hi ∞ A hi = mP [T A = m|X0 = i] m=1 ∞ = m pij P [T A = m − 1|X0 = j] m=1 j∈V ∞ = (m − 1)pij P [T A = m − 1|X0 = j] j∈V m=1 ∞ + pij P [T A = m − 1|X0 = j] j∈V m=1
- 36. ∞ (m − 1)pij P [T A = m − 1|X0 = j] j∈V m=1 ∞ = npij P [T A = n|X0 = j] j∈V n=1 = A pij hj j∈V
- 37. ∞ P [T A = m − 1|X0 = j] = 1 m=1 ∞ pij P [T A = m − 1|X0 = j] = pij = 1 j∈V m=1 j∈V A hi = A pij hj +1 A hi = 0, f or i ∈ A j∈V
- 38. w(i, k) w(k, j) pij = di dk k∈V2 hi (t + 1) = pij hj (t) + 1, hi (0) = 0 j=s
- 47. c(QT , U RL, U ser) p(QT |U RL, U ser) = Q c(Q, U RL, U ser) p(U RL|Q, U ser) ˆ
- 48. p(U RL|QT , U ser)p(QT |U ser) ˆ p(QT |U RL, U ser) = p(U RL|U ser) p(QT |U RL, U ser) k=j wik wij = 1 − p(QT |U RL, U ser) ˆ p 4 p(U RL|Q, IP ) = ˆ λi p(U RL|Q, IPi ) i=0
- 53. Generalized syntactic and semantic models of query reformulation (SIGIR 2010)
- 57. p(x, y) P M I(x, y) = log p(x)p(y) P M I(x, y) P M I(J)(x, y) = − log p(x, y) P M I(x, y) P M I(S)(x, y) = − log p(x) P M I(x, y) P M I(G)(x, y) = − log(p(y))
- 58. “Sorted”
- 60. Ns Nt Ns,t p(qs ) = , p(qt ) = , p(qs , qt ) = N N N P M IW eb (qs , qt )
- 61. dA (x, y) ∀a, b ∈ T, cE1 (a, b) = 1 if a = b ∀a, b ∈ T, cE2 (a, b) = dA (a, b) if a = N U LL||b = N U LL
- 63. s(wi , wj ) = 2 − 2f (wi , wj ) + (wi , wj ) ∀a, b ∈ T, cGE (a, b) = s(a, b) if a = N U LL||b = N U LL
- 64. qs = qs qt , qt = qt qs 1 if wi = wj ∧ wi ∈ qs ∧ wj ∈ qt ni,j (qs , qt ) = 1 if wi ∈ qs ∧ wj ∈ qt (|qs ||qt |) 0 otherwise Ni,j = ni,j (qs , qt ) qs ,qt
- 65. Ni,· = Ni,j , N·, = Ni,j , N = Ni,j j i i,j Ni,· N·,j Ni,j P (i, ·) = , P (·, j) = , P (i, j) = N N N
- 67. ∀xi S(yj |xi ) = 1, S(λ|ξ) = 0 yj ∈T ∪{λ}
- 68. m n+z G(z)n!z! p(y|x) = S(yk , xk ) (n + z)! z=max(0,m−n) x y k=1
- 74. Similarity Spearman NN 0.500 GenEdit(G) 0.380 Oommen-Kashyap 0.470 GenEdit(J) 0.365 DistSim 0.438 SortedEdit2 0.320 Mean all 0.435 SortedEdit1 0.314 SortedGenEdit(S) 0.429 PMIWeb(G) 0.283 SortedGenEdit(G) 0.428 Edit2 0.270 PMIWeb(S) 0.417 Edit1 0.252 PMIWeb(J) 0.409 Length-target(Char) 0.139 SortedGenEdit(J) 0.408 Length-target(Term) 0.112 GenEdit(S) 0.382 log-prob target -0.161
- 75. Similarity Function Spearman Mean all 0.386 NN 0.432 PMIWeb(G) 0.369 GenEdit(G) 0.424 PMIWeb(J) 0.330 SortedGenEdit(G) 0.419 DistSim 0.322 GenEdit(S) 0.414 Edit1 0.299 SortedGenEdit(S) 0.407 SortedEdit1 0.298 GenEdit(J) 0.402 Edit2 0.292 Oommen-Kashyap 0.391 SortedEdit2 0.288 SortedGenEdit(J) 0.391 PMIWeb(S) 0.264 log-prob target 0.114 Length- -0.036 target(Char) Length- -0.077 target(Term)
- 76. Similarity Measure QS1500 CC2000 Oommen-Kashyap 0.470* 0.391*(6) SortedGenEdit(S) 0.429* 0.407* (4) SortedGenEdit(G) 0.428* 0.419* (2) SortedGenEdit(J) 0.408 0.391* (7) GenEdit(S) 0.382 0.414* (3) GenEdit(G) 0.380 0.424* (1) GenEdit(J) 0.365 0.402* (5) SortedEdit2 0.320 0.288 (11) SortedEdit1 0.314 0.298 (9) Edit2 0.270 0.292 (10) Edit1 0.252 0.299 (8)
- 81. “