Dynamic Anticipation and Completion for Multi-Hop Reasoning over Sparse Knowledge Graph
30153Authors
nan
Abstract
AbstractMulti-hop reasoning has been widely studied in recent years to seek an effective and interpretable method for knowledge graph (KG) completion. Most previous reasoning methods are designed for dense KGs with enough paths between entities, but cannot work well on those sparse KGs that only contain sparse paths for reasoning. On the one hand, sparse KGs contain less information, which makes it difficult for the model to choose correct paths. On the other hand, the lack of evidential paths to target entities also makes the reasoning process difficult. To solve these problems, we propose a multi-hop reasoning model over sparse KGs, by applying novel dynamic anticipation and completion strategies: (1) The anticipation strategy utilizes the latent prediction of embedding-based models to make our model perform more potential path search over sparse KGs. (2) Based on the anticipation information, the completion strategy dynamically adds edges as additional actions during the path search, which further alleviates the sparseness problem of KGs. The experimental results on five datasets sampled from Freebase, NELL and Wikidata show that our method outperforms state-of-the-art baselines. Our codes and datasets can be obtained from https://github.com/THU-KEG/DacKGR.
Keywords
Full Content
Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 609–626, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 609 Condolence and Empathy in Online Communities Naitian Zhou Computer Science and Engineering University of Michigan naitian@umich.edu David Jurgens School of Information University of Michigan jurgens@umich.edu Abstract Offering condolence is a natural reaction to hearing someone’s distress. Individuals fre- quently express distress in social media, where some communities can provide support. How- ever, not all condolence is equal—trite re- sponses offer little actual support despite their good intentions. Here, we develop compu- tational tools to create a massive dataset of 11.4M expressions of distress and 2.8M cor- responding offerings of condolence in order to examine the dynamics of condolence on- line. Our study reveals widespread disparity in what types of distress receive supportive con- dolence rather than just engagement. Build- ing on studies from social psychology, we an- alyze the language of condolence and develop a new dataset for quantifying the empathy in a condolence using appraisal theory. Finally, we demonstrate that the features of condolence in- dividuals find most helpful online differ sub- stantially in their features from those seen in interpersonal settings. 1 Introduction Millions of individuals experience emotional dis- tress each year from diverse circumstances such as personal loss or abuse. After such experiences, peo- ple often turn to their social circle in social media to convey their experiences and seek out emotional support (Brubaker et al., 2012; Brubaker and Hayes, 2011; De Choudhury and Kiciman, 2017). Often, support comes in the form of condolence where individuals connect with the distressed person, and express forms of sympathy, empathy, advice, and social connection, among others (Burleson, 2003). However, not all expressions of distress receive emotional support, nor do all condolence mes- sages offer equal levels of support (Davidowitz and Myrick, 1984). Given the wide-spread use of social media for seeking social support, what makes for an effective supportive message? Here, we perform the first major study of condolence in social media, examining what type of distress individuals seek support for, what linguistics factors are more likely to elicit condolence, and what types of condolence viewed as more helpful. Distress and emotional support have long been explored in work in social psychology and coun- seling (Burleson et al., 2009; Rack et al., 2008), frequently around bereavement and helping victims of abuse. NLP works have only recently examined emotional support in online spaces for mental and physical health (Biyani et al., 2014; Navindgi et al., 2016; Wang et al., 2015) and in communities ori- ented around goals like weight-loss (Manikonda et al., 2014); however, these focus on the general concept of supportiveness. In this work, we exam- ine distress as a universal phenomenon—not just related to health and death—and examine the strate- gies and helpfulness of responses to this distress. This study aims to computationally identify mechanisms and strategies for delivering effective and impactful condolence on social media. Convey- ing condolence is often difficult for many people (Cameron …
Content truncated to 500 words
Labeled Text
AbstractMulti-hop reasoning has been widely studied in recent years to seek an effective and interpretable method for knowledge graph (KG) completion. Most previous reasoning methods are designed for dense KGs with enough paths between entities, but cannot work well on those sparse KGs that only contain sparse paths for reasoning. On the one hand, sparse KGs contain less information, which makes it difficult for the model to choose correct paths. On the other hand, the lack of evidential paths to target entities also makes the reasoning process difficult. To solve these problems, we propose a multi-hop reasoning model over sparse KGs, by applying novel dynamic anticipation and completion strategies: (1) The anticipation strategy utilizes the latent prediction of embedding-based models to make our model perform more potential path search over sparse KGs. (2) Based on the anticipation information, the completion strategy dynamically adds edges as additional actions during the path search, which further alleviates the sparseness problem of KGs. The experimental results on five datasets sampled from Freebase, NELL and Wikidata show that our method outperforms state-of-the-art baselines. Our codes and datasets can be obtained from
Publication Details
- Paper ID:
- 30153
- Year:
- 2020
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan