Sentiment Analysis Farrokh Alemi Ph D Wednesday April
- Slides: 35
Sentiment Analysis Farrokh Alemi, Ph. D. Wednesday, April 24, 2019
Sentiment Analysis Automated Analysis of Opinions
Scope of Analysis Review, Sentence, Phrase
Scope of Analysis The doctor was terrible; I could not understand him. The Review, Sentence, Phrase nurses were a different story. They were a pleasure to communicate to.
Methods of Analysis Keywords/Rule Based, Statistical
Methods of Analysis The doctor was terrible; I could not understand him. The Keywords/Rule Based, Statistical nurses were a different story. They were a pleasure to communicate to.
Methods of Analysis The doctor was terrible; I could not understand him. The Keywords/Rule Based, Statistical nurses were a different story. They were a pleasure to communicate to.
Methods of Analysis He knows nothing and was killing me Keywords/Rule Based, Statistical with poor advice. You are killing it.
Keywords Context Insensitive
Keywords Can’t Scale Up to Big Data To Do List 1 Oh So 2 Many 3 Words
Statistical Radical Improvements
Op Training ini on Classified Opinion l e Lab Word / Phrase Extraction Machine Learning Model
Op Training ini on Classified Opinion l e Lab Word / Phrase Extraction Machine Learning Model
Op Training ini on Classified Opinion l e Lab Word / Phrase Extraction Machine Learning Model
Op Training ini on Classified Opinion Word / Phrase Extraction Machine Learning Model l e Lab Op Prediction ini New Phrase on Word / Phrase Extraction Machine Learning Model Label
Last Mile Problem Misses Obvious Text Occasionally
4 Statistical Methods Maximum Likelihood Ratio
Likelihood Ratio
100 praises 50 Complaints I am happy with front desk Likelihood Ratio All I can say is that the breast surgery made me happy and young I am happy we stayed with this doctor This surgeon is knife happy
Last Mile Problem Can’t Learn from One Case
I am happy with front desk Last Mile Problem LR (Happy)=0. 17 LR(Knife)=1 All I can say is that the breast surgery made me happy and young I am happy we stayed with this doctor This surgeon is knife happy
4 Statistical Methods Longest Match & Maximum Likelihood Ratio
Longest Match & Maximum Likelihood I am happy with front desk All I can say is that the breast surgery made me happy and young I am happy we stayed with this doctor This surgeon is knife happy
Complaints n 2533 670 371 284 120 99 34 33 28 22 21 12 10 9 8 7 4 4 3 2 2 2 1 1 take 0 0 0 1 0 1 0 0 0 1 time 0 1 0 0 1 1 0 1 0 0 0 0 0 1 not 1 0 0 1 0 0 1 1 0 0 0 1 0 schedule 0 0 1 0 0 0 0 1 1 0 patients 0 0 1 0 0 0 1 0 1 1 0 close 0 0 0 0 1 0 0 0 1 0 0 0 together 0 0 0 0 0 0 1 New Sentence
ot Complaints n 2533 670 371 284 120 99 34 33 28 22 21 12 10 9 8 7 4 4 3 2 2 2 1 1 take 0 0 0 1 0 1 0 0 0 1 time 0 1 0 0 1 1 0 1 0 0 0 0 0 1 not 1 0 0 1 0 0 1 1 0 0 0 1 0 o N t h c at Mschedule 0 0 1 0 0 0 0 1 1 0 patients 0 0 1 0 0 0 1 0 1 1 0 close 0 0 0 0 1 0 0 0 1 0 0 0 together 0 0 0 0 0 0 1
Praise n 34 81 1369 1 236 2 4 3263 2 16 133 1 26 2 1628 2 1 105 55 0 take 0 0 0 0 0 time 0 0 0 0 1 1 1 not 0 0 0 0 1 1 1 1 0 0 0 schedule 0 0 1 1 1 0 0 0 1 1 0 0 1 patients 0 0 1 1 0 0 0 1 0 close 0 1 0 0 0 0 0 1 0 0 together 1 0 0 Match to: 1 not together 0 0 1 0 0 0
Praise & Complaint Praise 1369 676 236 133 560 137 81 105 83 n. Compl 371 284 120 99 34 33 28 22 21 take 0 1 0 0 0 1 time 0 0 1 1 0 not 0 0 0 1 schedule 0 0 1 0 0 0 patients 1 0 0 0 1 0 close 0 0 0 1 0 0 together 0 0 0 0 Match to: 0 take not
Likelihood Ratios Phrase together close patients not time take schedule Prevalence in Complaints Praise 10 34 28 81 371 1369 2533 3263 670 1628 284 676 120 236 LR 0. 59 0. 69 0. 54 1. 56 0. 83 0. 84 1. 02
Likelihood Ratios Prevalence in Phrase Complaints Praise close schedule 3 NULL patients schedule 2 4 together schedule NULL 2 together time NULL 2 close time NULL 1 close take NULL 1 patients take 7 43 schedule take 2 5 not take 21 83 time take 34 560 patients time 22 105 schedule time 12 55 not time 33 137 together not 4 2 close not 8 16 patients not 99 133 schedule not 9 26 close patients 2 1 LR 4. 00 1. 01 0. 33 0. 50 0. 33 0. 80 0. 51 0. 12 0. 44 0. 48 4. 02 1. 01 1. 50 0. 70 4. 02 Length of Match 2 2 2 2 2
Likelihood Ratios Phrase patients schedule not together time take patients time take close not take patients not take close time take schedule time take not time take patients schedule time patients not time together patients not time take patients schedule time take Prevalence in Complaints Praise 1 2 1 NULL 4 71 NULL 6 NULL 2 NULL 12 NULL 30 NULL 5 NULL 1 NULL 6 NULL 1 Length of LR Match 1. 01 3 2. 00 3 0. 11 3 0. 50 3 0. 14 3 0. 33 3 0. 08 3 0. 03 3 0. 17 3 0. 50 3 0. 14 4 Longest 0. 50 4 t s e ong g r t S on t m a es g n Lo
4 Statistical Methods Longest Match & Largest Switch in Sentiments
Longest Match & Largest Switch Competent staff but surgeon is knife happy
For the "Complaint" topic, "The doctor took a long time to talk to me" was classified as "False". We decided on this classification because of the combination of words doctor, took, time, talk. There were 60870 cases in the training set (14364 positive, 46506 negative). Prior odds of positive was set to 1. The following words were excluded as common words not central to the analysis: the, a, to, me. The index sentence had 5 remaining words. The minimum number of replications needed for a single word match was 96. The following combinations of words were examined before the process was aborted: Negative Likelihood Ratio Required Minimum Repetition Z 0 6 0. 14 0. 32 135. 90 291 493 1. 91 96 39. 49 Order Word Combination Positive 1 doctor, took, time, talk 2 long
The Way Forward? Not Sure?
- Chenghua lin
- Albizza
- Flora bölgeleri
- Cnler
- Protsta
- Monera alemi
- Analysis twitter mdausfishkinsparktoro
- W62vr
- Subjectivity in sentiment analysis
- Aspect based sentiment analysis using bert
- Bing liu sentiment analysis
- What is nvivo used for
- Sentiment analysis tools comparison
- Moontae lee
- Sentiment analysis for hotel reviews
- Turkish sentiment analysis
- Turkish sentiment analysis
- Azure twitter sentiment analysis
- Slide to doc.com
- Sentiment analysis with deep learning using bert
- Sentiment analysis conclusion
- Sentiment analysis
- Monday tuesday wednesday thursday friday calendar
- Wednesday evening prayer
- Wednesday seminar
- Ib grades to percentage
- Web analytics wednesday
- How to write wednesday
- Happy wednesday february
- Monday my favorite day
- Skinny wednesday
- Flmsg download
- Tuesday wednesday saturday
- Happy wednesday
- Happy wednesday march
- Monday tuesday wednesday thursday friday saturday sunday