Learning Prototypical Goal Activities for Locations Tianyu Jiang, - - PowerPoint PPT Presentation
Learning Prototypical Goal Activities for Locations Tianyu Jiang, Ellen Riloff University of Utah July 17, 2018 In Infer erenc ences es We e Are e Mak aking ing John went to the restaurant. What did he do? sat down
Tianyu Jiang, Ellen Riloff University of Utah July 17, 2018
2
John went to the restaurant. What did he do?
3
People go to __________ to __________ . We refer to an activity that represents a common reason why people typically go to a location as a prototypical goal activity (goal-act). libraries study churches pray hospitals see a doctor, have surgery ACL learn others’ work, meet people, give talk
4
Conversational systems, question answering, semantic disambiguation... Notice different inferences of “got”!
(a) She went to the kitchen and got chicken. (b) She went to the supermarket and got chicken. retrieve purchase
I just went to the beach. Did you go swimming?
5
6
7
(Loc, Act)
Activity Profile Y Ranked list of Activities for Locations
Until Convergence Location Similarity W Activity Similarity A
Corpus Labeled Data
Dataset: 2011 Spinn3r Weblog Subset (Burton et al., 2011) Pattern 1: go to X to Y extract (Loc, Act) pairs Pattern 2: Y in/at X extract more Acts
8
9
10
(Loc, Act)
Activity Profile Y Ranked list of Activities for Locations
Until Convergence Location Similarity W Activity Similarity A
Corpus Labeled Data
Activity profile matrix Y,
where Yi,j represents the strength of the j th activity aj being a goal-act for the i th location li.
11
a1=buy book a2=eat burger
...
am=pray l1=McDonald’s
5 300 1
l2=Burger King
2 500 2
l3=bookstore
400 20 4 ...
ln=church
5 10 700
An illustration of the activity profile matrix Y .
Intuitively, we assume that similar locations share similar activity profiles, which motivates the objective function over Y:
Initialization yi
0is a mix of co-occurrence data and labeled data.
12
13
(Loc, Act)
Activity Profile Y Ranked list of Activities for Locations
Until Convergence Location Similarity W Activity Similarity A
Corpus Labeled Data
14
co-occurrence word matching embedding
We use Amazon Mechanical Turk to ask workers to provide ONE primary activity that is the reason why a person would go to the listed locations. People go to LOC to ___ ___
VERB NOUN
We got answers for 200 locations from each of the 10 workers.
15
16
Location Gold Goal-Acts Toys R Us buy toys (9), browse gifts sink wash hands (7), wash dishes (3) airport catch flight (7), board planes, take airplane, take trips bookstore buy books (6), browse books (2), browse bestsellers, read book lake go fishing (3), go swimming (3), drive boat (2), ride boat, see scenery chiropractor get treatment (3), adjust backs (3), alleviate pain (2), get adjustment, get aligned Chinatown buy goods (2), buy duck, buy souvenirs, eat dim sum, eat rice, eat won- tons, find Chinese, speak Chinese, visit restaurants
Goal-acts provided by human annotators.
96% 78% 53% 39% 25% 15% 6% 2% 0.5% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2 3 4 5 6 7 8 9 10
% of Locations # of Annotators Listing the Same Activity
Percentage of locations that have at least one goal-act assigned by multiple annotators.
40%
At least half of annotators listed the same goal-act for nearly 40% of the locations.
1
Only 1 location was assigned exactly the same goal-act by all annotators. 17
Our systems produce a ranked list of hypothesized goal-acts for a location. Mean Reciprocal Rank(MRR) is used to judge the quality of the top 10 activities for each location.
18
19
MRRE MRRP EMBED
0.02 0.09
PMI
0.20 0.33
FREQ
0.23 0.34
AP
0.28 0.38
AP+AL
0.28 0.40
AP+AO
0.23 0.33
AP+AE
0.25 0.36
AP+AL+E
0.29 0.42
Scores for MRR.
20
TOP1 TOP2 TOP3 EMBED
0.05 0.08 0.12
PMI
0.25 0.36 0.41
FREQ
0.23 0.32 0.40
AP
0.29 0.41 0.47
AP+AL
0.32 0.44 0.49
AP+AO
0.24 0.35 0.43
AP+AE
0.28 0.40 0.47
AP+AL+E
0.35 0.44 0.52
Scores for Top K results.
21
Location Gold Activity List AP + AL+E Top 3 PMI Top 3
bookstore buy book (6) browse book (2) browse bestseller read book buy book purchase book see book buy copy purchase book buy book pharmacy get drug (4) fill prescription (3) get prescription (2) buy medicine find medicine get prescription pick up prescription buy pill fill prescription pick up prescription university get degree (4) gain education (5) watch sport gain education further education gain knowledge study law study psychology pursue study Meijer buy grocery (8) buy cream
buy item go shopping get item check out deal have shopping post today Examples of Top 3 hypothesized prototypical goal activities.
22 Examples of Top 3 hypothesized prototypical goal activities.
Location Gold Activity List AP + AL+E Top 3 PMI Top 3
market buy grocery (6) buy fresh, buy goods buy shirt, find produce make money eat out eat lunch have demand increase competition lead player phone make call (4), NOT LOC (2) answer call, call friend have conversation stop ring play game browse website view website put number have number put card
locations.
goal-acts with locations.
knowledge.
23
24