Speech Project
Week 2 Report
b02901085 徐瑞陽
b02901054 方為
March 1, 2016
Task 5:
Aspect Based Sentiment Analysis (ABSA)
We submitted results for slots 1 and 3 of subtask 1
Subtask 1: Sentence-level ABSA
Given a review text about a target entity (laptop, restaurant, etc.),
identify the following information:
-
Slot 1: Aspect Category
- ex. ''It is extremely portable and easily connects to WIFI at the library and elsewhere''
----->{LAPTOP#PORTABILITY}, {LAPTOP#CONNECTIVITY}
- ex. ''It is extremely portable and easily connects to WIFI at the library and elsewhere''
-
Slot 3: Sentiment Polarity
- label: (positive, negative, or neutral)
Slot1 : Aspect Detection
Subtask1: Sentence-level
Slot1 : Aspect Detection
| Restaurant | Laptop | |
|---|---|---|
| Team | 12-class F-Measure |
81-class F-Measure |
| Our Results (SVM) | 65.455 (15th overall, 6th in Constrained) | 43.754 (13th overall, 5th in Constrained) |
| Unconstrained Best | 73.031 (1st overall) | 51.937 (1st overall) |
| Constrained Best | 71.494 (5th overall) | 47.891 (5th overall) |
Results
Slot3 : Polarity Detection
Subtask1: Sentence-level
Slot3 : Polarity Identification
Chain-structured LSTM vs Tree-structured LSTM
Slot3 : Polarity Identification
Proposed Model
Slot3 :
Polarity Detection
Results
| Restaurant | Laptop | |
|---|---|---|
| Team | 3-Class Accuracy | 3-Class Accuracy |
| Our Submission | 78.114 (20th overall, 8th in Constrained) | 75.905 (7th overall, 1st in Constrained) |
| Unconstrained Best | 86.729 (2nd overall) | 82.772 (1st overall) |
| Constrained Best (or next best) | 88.126 (1st overall) | 74.282 (9th overall) |
We submitted our results under "Constrained", but we are not sure if our run belongs in this category because we used GloVe vectors
Future Work
Framework
Tree Nodes
Implementation
- Task
- SemEval: ABSA
- bAbI
- Library
- Theano or TensorFlow