File size: 11,169 Bytes
fad35ef
1
{"forum": "SygLHbcapm", "submission_url": "https://openreview.net/forum?id=SygLHbcapm", "submission_content": {"title": "Integrating User Feedback under Identity Uncertainty in Knowledge Base Construction", "authors": ["Ari Kobren", "Nicholas Monath", "Andrew McCallum"], "authorids": ["akobren@cs.umass.edu", "nmonath@cs.umass.edu", "mccallum@cs.umass.edu"], "keywords": ["user feedback", "entity resolution", "identity uncertainty"], "TL;DR": "This paper develops a framework for integrating user feedback under identity uncertainty in knowledge bases. ", "abstract": "Users have tremendous potential to aid in the construction and maintenance of knowledges bases (KBs) through the contribution of feedback that identifies incorrect and missing entity attributes and relations. However, as new data is added to the KB, the KB entities, which are constructed by running entity resolution (ER), can change, rendering the intended targets of user feedback unknown\u2013a problem we term identity uncertainty. In this work, we present a framework for integrating user feedback into KBs in the presence of identity uncertainty. Our approach is based on having user feedback participate alongside mentions in ER. We propose a specific representation of user feedback as feedback mentions and introduce a new online algorithm for integrating these mentions into an existing KB. In experiments, we demonstrate that our proposed approach outperforms the baselines in 70% of experimental conditions.", "pdf": "/pdf/7c4a62f427867407882e908a39323c40a0086210.pdf", "archival status": "Archival", "subject areas": ["Information Integration", "Human computation"], "paperhash": "kobren|integrating_user_feedback_under_identity_uncertainty_in_knowledge_base_construction", "_bibtex": "@inproceedings{\nkobren2019integrating,\ntitle={Integrating User Feedback under Identity Uncertainty in Knowledge Base Construction},\nauthor={Ari Kobren and Nicholas Monath and Andrew McCallum},\nbooktitle={Automated Knowledge Base Construction (AKBC)},\nyear={2019},\nurl={https://openreview.net/forum?id=SygLHbcapm}\n}"}, "submission_cdate": 1542459709870, "submission_tcdate": 1542459709870, "submission_tmdate": 1580939656766, "submission_ddate": null, "review_id": ["Bye0ls1mfV", "SylJd4urG4", "BkeLdizyEN"], "review_url": ["https://openreview.net/forum?id=SygLHbcapm&noteId=Bye0ls1mfV", "https://openreview.net/forum?id=SygLHbcapm&noteId=SylJd4urG4", "https://openreview.net/forum?id=SygLHbcapm&noteId=BkeLdizyEN"], "review_cdate": [1547004662037, 1547170919139, 1548852078404], "review_tcdate": [1547004662037, 1547170919139, 1548852078404], "review_tmdate": [1550269659669, 1550269631752, 1550269627104], "review_readers": [["everyone"], ["everyone"], ["everyone"]], "review_writers": [["AKBC.ws/2019/Conference"], ["AKBC.ws/2019/Conference"], ["AKBC.ws/2019/Conference"]], "review_reply_count": [{"replyCount": 0}, {"replyCount": 0}, {"replyCount": 0}], "review_replyto": ["SygLHbcapm", "SygLHbcapm", "SygLHbcapm"], "review_content": [{"title": "A nice algorithmic contribution for integrating user feedback for KBC", "review": "This paper introduces a method to integrate user feedback into KBs in the presence of identity uncertainty, a problem that arises in the integration of new data in knowledge base construction. The proposed method represents user feedback as feedback mentions and uses an online algorithm for integrating these mentions into the KB.\n\nThe paper targets an important problem in knowledge base construction, i.e., integrating user feedback in the online setting. The proposed hierarchical model looks reasonable and effective. And overall, the work is well presented. \n\nThe paper makes an algorithmic contribution. The contribution is, however, limited from the perspective of human computation. The experiment uses simulated user feedback for evaluating the method. In real-world settings, user feedback can be skewed to certain types (e.g., negative feedback) or be noisy (so the feedback is not reliable). How would these affect the result?", "rating": "7: Good paper, accept", "confidence": "3: The reviewer is fairly confident that the evaluation is correct"}, {"title": "lack of details, few novelty", "review": "This paper presents a hierarchical framework for integrating user feedback for KB construction under identity uncertainty.\n\n1. it is unclear about the algorithm implementation, such as what is the implementation of feedback mention.\n\n2. There is a definition about attribute map, but I can't find where the model uses it. Same thing for the precision of a node pair.\n\n3. How to calculate the function g(.) is unclear either. \n\n4. In section 5.2, the COMPLETE definition seems not correct, it is still the definition of PURE.\n\n5. The example for constructing positve/negative feedback is too vague.\n\n6. The experiment section needs more analysis including qualitative result.", "rating": "3: Clear rejection", "confidence": "4: The reviewer is confident but not absolutely certain that the evaluation is correct"}, {"title": "Relevant contribution for humans-in-the-loop in KB construction", "review": "The paper presents a novel solution to an interesting problem - when KBs are automatically expanded user feedback is crucial to identify incorrect and missing entity attributes and relations. More specifically, in the case of entity identity uncertainty, enabling the user feedback as part of the entity resolution mentions, appears novel and important. \nThe paper is well written and organized. \n\nPoints for improvement:\n- It would be interesting to see more in-depth analysis on examples where the proposed approach fails, and based on this to also outline open issues and future work. \n- The human computation aspects of the paper are lacking sufficient explanation in terms of implementation in real settings, as well as positioning with related work in human computation research\n- it would be interesting to know, what is the experimental design that authors would consider for an evaluation with actual user feedback vs. the simulated one", "rating": "7: Good paper, accept", "confidence": "4: The reviewer is confident but not absolutely certain that the evaluation is correct"}], "comment_id": ["BJx2OdWf74", "Hkl94_WMQE"], "comment_cdate": [1547995251567, 1547995186367], "comment_tcdate": [1547995251567, 1547995186367], "comment_tmdate": [1547995251567, 1547995186367], "comment_readers": [["everyone"], ["everyone"]], "comment_writers": [["AKBC.ws/2019/Conference/Paper51/Authors", "AKBC.ws/2019/Conference"], ["AKBC.ws/2019/Conference/Paper51/Authors", "AKBC.ws/2019/Conference"]], "comment_reply_content": [{"replyCount": 0}, {"replyCount": 0}], "comment_content": [{"title": "Experiment with mixed detailed and concise feedback", "comment": "Thank you for your positive review! \nWe would be happy to expand on our experimental results. While it is not possible for us to carry out a study with real human feedback (which requires IRB approval, etc.) we can perform experiments with a mixture of detailed and concise feedback and with different proportions of positive and negative feedback. Would this address your criticism (at least in part)?\n"}, {"title": "Details are included in the paper", "comment": "Thank you for your review. \nThe details you mention do appear in our paper, but perhaps they can be further clarified.  We address each of your criticisms below:\n\n1. The implementation of feedback mentions is described in the third paragraph of section 2.3: \u201cFMs are composed of two attribute maps\u201d the first called \u201cpackaging\u201d and the second called \u201cpayload\u201d.  Immediately after this description, we provide intuition for why both attribute maps are needed. In section 3.1.1, we describe precisely how both of these attribute maps are used during canonicalization.\n\n2. The model, g, computes the linkage score between two nodes using their attribute maps.  This is described in the second paragraph in 3.1.1: \u201cSince g scores the compatibility of two nodes based on their attribute maps...\u201d. Perhaps we can further clarify this in the paper by including some text describing the linkage function g more precisely. \n\nThe precision of a node pair is used during training. In the first paragraph of section 4: \u201cWe train g to regress to the precision\u2026\u201d Thus, during learning we run agglomerative clustering and, at each merge, train g to predict the precision of the corresponding node pair.\n\n3. The function g is computed from the attribute maps of the corresponding nodes (which differ by dataset).  It is true that we omit the specific features that we use, for example, whether two nodes have the same title with a corresponding positive count in their respective attribute maps, or whether one node contains an attribute A with a positive count and the second node contains the attribute A with a negative count.  We are open to including these features in a supplement.\n\n4. We agree that the definition for complete could be revised to be clearer. We will modify the definition as follows:\nA node v is complete if for some $i$,\n\\lvs{e^\\star_i} \\subseteq lvs{v}.\nThus, if for some $i$, the leaves of $e^\\star_i$ are a subset of the leaves of $v$, then $v$ is complete. We note that the definition for complete is described in words immediately after the mathematical statement in the paper and is different from the definition of pure.\n\n5. What part of the detailed/concise feedback needs further explanation? As described in 5.2.1 and 5.2.2 and as detailed in the figure, the simulation procedure finds a \u201cdestination\u201d and \u201ctarget\u201d for each edit.  The simulated edit contains a packaging that includes all of the positive attributes at the destination and a payload containing the attributes at the target (either with positive or negative weight, as described in the paper). We\u2019re happy to describe the feedback generation process in more detail.\n\n6. We believe that qualitative results would not be very illuminating, but we\u2019re open to including some. Would you find informative an  example of an edit added to a tree and the subsequent modification of inferred entities?\n"}], "comment_replyto": ["Bye0ls1mfV", "SylJd4urG4"], "comment_url": ["https://openreview.net/forum?id=SygLHbcapm&noteId=BJx2OdWf74", "https://openreview.net/forum?id=SygLHbcapm&noteId=Hkl94_WMQE"], "meta_review_cdate": 1549982240517, "meta_review_tcdate": 1549982240517, "meta_review_tmdate": 1551128218505, "meta_review_ddate ": null, "meta_review_title": "Interesting topic to be presented", "meta_review_metareview": "The paper presents an interesting methodology. The results are interesting, however the paper really misses out on an in-depth discussion and reflection of the pros and cons of this approach as well as on a proper related work comparison to similar approaches. ", "meta_review_readers": ["everyone"], "meta_review_writers": [], "meta_review_reply_count": {"replyCount": 0}, "meta_review_url": ["https://openreview.net/forum?id=SygLHbcapm&noteId=rkFXcIgBE"], "decision": "Accept (Poster)"}