task-2

System

• golf_system_0: using word embedding to construct document vector and calculate similarity by four features including cosine sim and euclidean distance.

• golf_system_1: introducing TF-IDF cosine similarits (tweet VS vclaim_content, tweet VS vclaim_title) and BM25 scores (tweet VS vclaim_content, tweet VS vclaim_title) as features, Linear SVC as model

• golf_system_2: introducing TF-IDF cosine similarits (tweet VS vclaim_content, tweet VS vclaim_title) and BM25 scores (tweet VS vclaim_content, tweet VS vclaim_title) as features, Logistic Regression as model

• golf_system_3: introducing TF-IDF cosine similarits (tweet VS vclaim_content, tweet VS vclaim_title) and BM25 scores (tweet VS vclaim_content, tweet VS vclaim_title) as features, Linear Regression as model

Result

BaseLine:

metric	@depth	score
map	1	0.470
map	3	0.601
map	5	0.609
map	10	0.615
map	20	0.617
map	all	0.619
precision	1	0.472
precision	3	0.249
precision	5	0.156
precision	10	0.082
precision	20	0.043
precision	all	0.000
reciprocal_rank	1	0.472
reciprocal_rank	3	0.603
reciprocal_rank	5	0.611
reciprocal_rank	10	0.616
reciprocal_rank	20	0.619
reciprocal_rank	all	0.621

System 1:

metric	@depth	score
map	1	0.622
map	3	0.622
map	5	0.622
map	10	0.622
map	20	0.622
map	all	0.622
precision	1	0.624
precision	3	0.208
precision	5	0.125
precision	10	0.062
precision	20	0.031
precision	all	0.000
reciprocal_rank	1	0.624
reciprocal_rank	3	0.624
reciprocal_rank	5	0.624
reciprocal_rank	10	0.624
reciprocal_rank	20	0.624
reciprocal_rank	all	0.624

System 2:

metric	@depth	score
map	1	0.612
map	3	0.612
map	5	0.612
map	10	0.612
map	20	0.612
map	all	0.612
precision	1	0.614
precision	3	0.205
precision	5	0.123
precision	10	0.061
precision	20	0.031
precision	all	0.000
reciprocal_rank	1	0.614
reciprocal_rank	3	0.614
reciprocal_rank	5	0.614
reciprocal_rank	10	0.614
reciprocal_rank	20	0.614
reciprocal_rank	all	0.614

System 3: map@5 0.586