[Talk Summary 5] Two Case Studies in Semantic Inference


Dipanjan Das from Google presented the talk "Two Case Studies in Semantic Inference" on October, 14, 2016. He performed the two different semantic inference tasks. The first one focuses on the structure of natural language questions. And, the other is about more unstructured forms.

Firstly, Dipanjan described a method for parsing natural language questions to logical forms. These logical forms can be softly mapped to the information stored in the structured knowledge base. And then the system matches the forms (sub-graphs) from knowledge base to Question Answer mechanism. In the semantic parsing process, he presented a Dependency Parser that extract sentences' structure, and DepLambda to parse into logical forms, which is based on lambda calculus. 

Dipanjan carried out an empirical study to compares DepLambda techniques to other baselines. He showed that DepLambda, in two test collections, performs better than Simple Graph, CCG Graph and Deptree.

In the second part of the talk, Dipanjan shared the motivation about finding relevant sentences without representing them into a specific model such as vector model with heavy weight. He presented two semantic inference tasks for more free forms of language to calculate the similarity between sentences. The idea is a simple attention-based approach to text similarity that is trivially parallelizable. The first task he used decoder recurrent neural network, and applied decomposable attention in the second task.

For example, the best matches of two sentences "in the park alice plays a flute solo" and "someone playing music outside" are (park, outside), (alice, someone), (play, playing), and (flute, music).

The empirical results on Stanford Natural Language Inference Corpus showed that although with a simple approach without global sentence-level representations, his systems received the better results compared to other top systems.

NOTES: you can google these two sentence and see the results
"oreo hard drugs study"
"toyota prius how to operate gas tank"

10/14/2016
100 Porter Hall
Carnegie Mellon University

Comments

Post a Comment

Popular posts from this blog

Personalized Access to Scientific Publications: from Recommendation to Explanation

Dario De Nart, Felice Ferrara, and Carlo Tasso ( UMAP-2013 ) http://link.springer.com/chapter/10.1007/978-3-642-38844-6_26 In the paper “Personalized Access to Scientific Publications: from Recommendation to Explanation”, Dario De Nart et al. present a Recommender and Explanation System (RES) that is able to recommend scientific publication and show explanation to users.  In order to do recommendation, Dario De Nart et al., firstly, use the The Dikpe Keyphrase extraction algorithm introduced by Pudota (2010) to extract keyphrases from papers. Each keyphrase has a weight called keyphraseness that reveals the several lexical and statistical indicators exploited in the extraction process. Higher is the keyphraseness, more relevant is the KP. For each paper in the collection, they represent it as a conceptual graph. Each node in the graph is a term broken down from a KP. Two nodes are connected when they appear in the same KP. For instance, the KP “document retrieval” is split into tw
Read more

[Talk Summary 2] Dynamic Information Retrieval Model

Image
In this seminar, prof essor Grace Hui Yang presents a dynamic information retrieval model which helps users explore the information space in order to find out which documents are relevant and which aren't, satisfying their information need. The dynamic IR task aims to find relevant documents for a session of multiple queries. It happens when information needs are complex, vague, evolving, often containing multiple subtopics. A dynamic system is one which changes or adapts over time, based on a sequence of events. While static IR does not learn directly from users and the parameters are updated periodically, dynamic IR is developed from interactive IR which exploits users' feedback to give recommendation or improve the search result or optimize ranking, which is called dynamic ranking principle. But interactive IR just treat interaction independently and response to immediate feedback. Dynamic IR tries to optimize over all interaction, uses it for long term gain and models
Read more

FolkTrails: Interpreting Navigation Behavior in a Social Tagging System

Thomas Niebler, Martin Becker, Daniel Zoller, Stephan Doerfel, & Andreas Hotho ( CIKM-2016 ) http://dl.acm.org/citation.cfm?doid=2983323.2983686 In the paper “FolkTrails: Interpreting Navigation Behavior in a Social Tagging System”, Niebler et al. present an investigate navigation trails in the popular scholarly social tagging system BibSonomy. They tried to understand users’ dynamic browsing behaviors on this kind of systems, how individual users behave differently on navigation, and whether the semantic nature of the underlying folksonomy can help to explain navigation. Firstly, Niebler et al. study different hypotheses about the navigational user behavior in Bibsonomy, a social bookmarking system. They test these hypotheses by using HypTrails, an approach utilizing first-order Markov chain models and Bayesian inference for expressing and comparing hypotheses about human trails. These hypotheses include: Uniform hypothesis: serves as a baseline, users randomly pick a l
Read more