An Open Relation Extraction System for Web Text Information

Round 1

Reviewer 1 Report

The topic and the paper are interesting, but there are a few significant drawbacks:

• There are numerous grammar and spelling errors.
• Some parts of the text are repeated (e.g., lines 296-301 on page 8 are immediately repeated as lines 302-305)
• Some formulas are unclear (e.g., what exactly is K_Inject in Eq. (20)), and others are missing (Eq. (7) is a repetition of Eq. (5) and not the formula that should be there)
• There is an overall lack of attention to detail, e.g., names of LaTeX packages in the text (c.f. lines 498, 529), or typesetting variable names in a standard font without using subscripts.

The proposed methods clearly cannot beat the state of the art for >10-shot learning. It thus would be beneficial to the readers to extend the discussion on the results with a very limited number of shots.

Author Response

Thanks for your sincere review and kindness! Our revisions are as follows:

1. We have thoroughly checked and corrected the grammar and spelling errors. Furthermore, we modify the paper for appropriate expressions. 
2. Maybe you mean that part lines 302-305(304-307) is repeated in part lines 309-311 on page 8. We have checked and deleted it. The part lines 296-301 might isn't repeated as lines 302-305(308-311). The former part is the threshold setting, for instance selector, and the latter is the threshold setting for the relation classifier. And we have revised the more detailed nomenclature for better reading.
3. We have revised the formula(7).
4. For a better reading, we supplemented the function of K_inject.
5. For the problem of LaTeX packages, we have revised parts lines 498 and 529. And we have typeset the LaTeX text.

Low-resource relation learning is a strength of ORES. It is more conducive to studying the model's generalization ability, which is also the direction we will continue to study. Thanks for your comments！

Author Response File: Author Response.pdf

Reviewer 2 Report

An interesting paper but  in need of a thorough review of the English and presentation.

Line 13: Text records 80 percent of the information of human civilization.

Reference?

 Line 89-90: As shown in Figure 1, given some seed instances with new relation facts and few sample data.

The paper generally needs a first-language English speaker to help with grammar and sentence construction.

 Line 270: Figure 3 requires explanation. This is actually given a few lines later but refers to Figure 4.

Line 278: As shown in Figure  4

Appears to refer to Figure 3.

Line 284 and elsewhere. Si, ENT and so on, should be in mathematical notation.

Line 288: (c) Load sentences in the Coarse selection set into sample selector 1.

There is no ‘sample selector 1’ in the diagram.

The rest of the paper requires similar rewriting to make these interesting results suitable  for publication.

 

Author Response

Thanks for your sincere review! Our revisions are as follows:

1. The reference is related to statistics in the social sciences domain. So we have not included it as a reference. Low-resource relation learning is a strength of ORES. It is more conducive to studying the model's generalization ability, which is also the direction we will continue to study. The reference list is as follows： https://breakthroughanalysis.com/2008/08/01/unstructured-data-and-the-80-percent-rule/
2. We have revised the more appropriate formulation for lines 89-90. We checked and corrected grammar and spelling errors. For a better reading, we modified the more appropriate presentation.
3. We have supplemented the explanation in Figure 3.
4. Line 278: As shown in Figure  4. Figure 4 shows the process of instance selector in operating sentences. Your review is more appropriate. We revised it in detail.
5. Lines 284(287). We have revised Si and Ent to be mathematical notations. Furthermore, we thoroughly checked all remaining parts and revised them to the canonical mathematical notation.
6. Line 288. There is no ‘sample selector 1’ in the diagram. Our representation is not accurate enough. In order to make readers read better, we uniformly modify the sample selector to the instance selector in this paper.

Low-resource relation learning is a strength of ORES. It is more conducive to studying the model's generalization ability, which is also the direction we will continue to study. Thanks for your patient review in detail.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Thank you for improving the presentation of your work.