Course talk:CPSC522/Generic Aspect-based Aggregation of Sentiments

Hi Prithu,

Good job. The content of this page is great and I learned a lot. But I think the hierarchy or layout of this page need be redesigned. Here are my suggestions:

• Experiments and Training Section should not be under Hypothesis Section. From my perspective, Hypothesis Section should be concise and short and Experiments and Training can be put in a implementation section.
• Conclusion Section is missing.
• Builds on should be under Abstract Section like the template page.
• It seems that you have not still finished Background Knowledge Section.

Sincerely,

Ke Dai

Thanks for your feedback Kedai. Experiments and Training Section should not be under Hypothesis Section. From my perspective, Hypothesis Section should be concise and shot and Experiments and Training can be put in a implementation section.

Not sure what you mean. I saw experiment as an empirical validation of my hypothesis. And found it easy to link in that way. Hope that keeps the flow alright

Conclusion Section is missing.

I have a discussion section instead :)

It seems that you have not still finished Background Knowledge Section.

Should be alright. Can you make sure you are looking at the latest snapshot.

best, Prithu

Hi Prithu,
Great page! I really enjoyed reading it. I have worked on ontology and sentiment analysis, so I found this page really interesting. Some questions and suggestions I had:

• In the Sentiment analysis vs Sentiment Aggregation section SVM has been mentioned. But the full form in mentioned in related work section. I think you can mention the full form in the earlier section.
• The ontology figure needs to be a bit bigger to improve readability. You can also label it figure 1 as you are referring to it in the text.
• It would be great if you could provide a link for ConceptNet?
• MLE and EM mentioned in bayesian inference mean Maximum Likelihood And Expectation Maximization?
• In the aspect sentiment annotation section who is bing liu? The figure needs some explanation.
• I was not able to find the restaurant ontology when I clicked on the link.
  

The document name you requested (/ontologies/restaurant.owl|Restaurant) could not be found on this server. However, we found documents with names similar to the one you requested. Available documents: /ontologies/restaurant.owl (common basename) Is it this one? http://wise.vub.ac.be/ontologies/restaurant.owl

• Same thing with the reviews.: Document not found
• I did not understand this line "We also compare with another variant where we augement tree-models the influence learning, in place of strict hierarchy driven weighing." Should it be "augment tree models with influence learning"?
• In the training section who is Next?
• In the example arff file it has been mentioned that you are limiting to 10 relations. By relations do you mean attributes(but there are only 7 attributes in the figure)?

Samprity Thanks a lot for detailed and valuable feedback. Response inline: Your feedback really helped me a lot

Hi Prithu,
Great page! I really enjoyed reading it. I have worked on ontology and sentiment analysis, so I found this page really interesting. Some questions and suggestions I had:

• In the Sentiment analysis vs Sentiment Aggregation section SVM has been mentioned. But the full form in mentioned in related work section. I think you can mention the full form in the earlier section.

Corrected

• The ontology figure needs to be a bit bigger to improve readability. You can also label it figure 1 as you are referring to it in the text.

Labelling added. For closer look please click on the figure. The figure is big, no matter how large i make, some nodes would still remain hard to read :)

• It would be great if you could provide a link for ConceptNet?

WordNet link added

• MLE and EM mentioned in bayesian inference mean Maximum Likelihood And Expectation Maximization?

Yes, I added pointers too now

• In the aspect sentiment annotation section who is bing liu? The figure needs some explanation.

Ya sorry my bad. Clarified that part

• I was not able to find the restaurant ontology when I clicked on the link.
  

Links are all fixed now The document name you requested (/ontologies/restaurant.owl|Restaurant) could not be found on this server. However, we found documents with names similar to the one you requested. Available documents: /ontologies/restaurant.owl (common basename) Is it this one? http://wise.vub.ac.be/ontologies/restaurant.owl

• Same thing with the reviews.: Document not found
• I did not understand this line "We also compare with another variant where we augement tree-models the influence learning, in place of strict hierarchy driven weighing." Should it be "augment tree models with influence learning"?

Yes. Essentially vanilla tree is anyway bad. So next was to add learning on tree which did well

• In the training section who is Next?

I am not sure what you mean. There is none "named" next in that section.

• In the example arff file it has been mentioned that you are limiting to 10 relations. By relations do you mean attributes(but there are only 7 attributes in the figure)?

Ya i removed them from figure. Figure and arff do not correspond, so don't match them.

Hi Prithu
Thanks for the clarifications! I must have read "proposed by. Next" as "proposed by Next"!!

Hi Prithu,

I enjoy reading this page. Just a few suggestions:

One quick proofreading might help.

>Generic Aspect-based Aggregation of Sentiments

You may want to add a sentence about the page in the first part of this section.

“Builds on” should not be in separate words. It was better to mention that in structured sentences.

Bayesian inference should be explained in the background section, not in the proposed framework section.

“lexicons provided by bing liu” needs citation.

Figures do not have numbers.

>Aspect Sentiment Annotation

The figure in this part is not explained enough. There are some parts than need to be explained more.

“work of Mukherjee et al.” needs citation.

You have explained your method very good. But at one glance I can say the texts in this page are too much. You may want to summarize the explanations about the motivations and the background section.

Bahare thanks for the suggestation and proof reading u did for the betterment of the page. Here are my responses:

Hi Prithu,

I enjoy reading this page. Just a few suggestions:

One quick proofreading might help. Yup >Generic Aspect-based Aggregation of Sentiments

You may want to add a sentence about the page in the first part of this section. Ok “Builds on” should not be in separate words. It was better to mention that in structured sentences. Sure I added that Bayesian inference should be explained in the background section, not in the proposed framework section. Not sure. As in this secction i dnt introduce general bayesian learning, as that would take even more words, for what you already complained. Rather I explain how i use it in my required settings. Hope this makes sense “lexicons provided by bing liu” needs citation. right Figures do not have numbers. added >Aspect Sentiment Annotation

The figure in this part is not explained enough. There are some parts than need to be explained more. I fixed it with a less confusing (hopefully :-)) figure now

“work of Mukherjee et al.” needs citation. True You have explained your method very good. But at one glance I can say the texts in this page are too much. You may want to summarize the explanations about the motivations and the background section. Tried a precise version now.

Hi Prithu,

Impressive work! Here is my feedback:

General

• Some proofreading would be good, but in general the grammar issues don't impact readability.

Ontology and Aggregation

• You state that two concepts are depicted in the figure, yet there are many more than that; it may confuse some readers.
• You state "dish has an influence for restaurant, following the opposite direction in the ontology connecting them", but the arrow goes from dish to restaurant; wouldn't that be the same direction?

Proposed framework

• Regarding the annotation of overall review sentiment - is that annotation performed by humans? You have a human graphic on the model but say you use a sentiment analyzer, so I'm not sure where the automated part ends and the human-annotated part begins.

Aspect Sentiment Annotation

• Who is Bing Liu?

Results

• Using bold font for the "winning" values would make the results tables easier to read.

Thanks a lot Jordon for helping reviews. Response in-line and I am sorry for many typos and silly errors Hi Prithu,

Impressive work! Here is my feedback:

General

• Some proofreading would be good, but in general the grammar issues don't impact readability.

Ontology and Aggregation

• You state that two concepts are depicted in the figure, yet there are many more than that; it may confuse some readers.
• You state "dish has an influence for restaurant, following the opposite direction in the ontology connecting them", but the arrow goes from dish to restaurant; wouldn't that be the same direction?

Fixed the whole business with a better figure. Hope that would help Proposed framework

• Regarding the annotation of overall review sentiment - is that annotation performed by humans? You have a human graphic on the model but say you use a sentiment analyzer, so I'm not sure where the automated part ends and the human-annotated part begins.

Ya there are two parts. First to identify sentiment at each sentence, this is automated by analyser. However for experiment we need ground truth on aggregated sentiment. This is where human intervention needed. I added a section explaining it. Hope that would clarify Aspect Sentiment Annotation

• Who is Bing Liu?

Cited now :-). He is not as popular as me, I forgot that ;-) Results

• Using bold font for the "winning" values would make the results tables easier to read.

Ya. Will do the painful but useful type-setting :D