LSI
Test results
- Topic cosine distribution (in logarithmic scale)
- Industry cosine distribution (in logarithmic scale)
- Results using Reuters' topic classifications — manual picks vs. angle cosine above:
- 1 standard deviation
- 2 standard deviations
- 3 standard deviations
- 4 standard deviations (unavailable, would exceed Google App Engine's quota)
- 5 standard deviations (unavailable, would exceed Google App Engine's quota)
- 6 standard deviations (unavailable, would exceed Google App Engine's quota)
- 7 standard deviations (unavailable, would exceed Google App Engine's quota)
- 8 standard deviations (unavailable, would exceed Google App Engine's quota)
- 9 standard deviations (unavailable, would exceed Google App Engine's quota)
- 10 standard deviations
- Results using Reuters' industry classifications — manual picks vs. angle cosine above:
- 1 standard deviation
- 2 standard deviations
- 3 standard deviations
- 4 standard deviations (unavailable, would exceed Google App Engine's quota)
- 5 standard deviations (unavailable, would exceed Google App Engine's quota)
- 6 standard deviations (unavailable, would exceed Google App Engine's quota)
- 7 standard deviations (unavailable, would exceed Google App Engine's quota)
- 8 standard deviations (unavailable, would exceed Google App Engine's quota)
- 9 standard deviations (unavailable, would exceed Google App Engine's quota)
- 10 standard deviations
- System's precision and recall — Averages against the number of dimensions and the decision threshold
- System's precision and recall — Best results for topic classifications
- System's precision and recall — Best results for industry classifications
- System's precision and recall, for topic classifications, using a decision threshold of:
- 1 standard deviation
- 2 standard deviations
- 3 standard deviations
- 4 standard deviations (unavailable, would exceed Google App Engine's quota)
- 5 standard deviations (unavailable, would exceed Google App Engine's quota)
- 6 standard deviations (unavailable, would exceed Google App Engine's quota)
- 7 standard deviations (unavailable, would exceed Google App Engine's quota)
- 8 standard deviations (unavailable, would exceed Google App Engine's quota)
- 9 standard deviations (unavailable, would exceed Google App Engine's quota)
- 10 standard deviations
- System's precision and recall, for industry classifications, using a decision threshold of: