Deep Learning for Entity Matching : A Design Space Exploration


Authors :
Sidharth Mudgal, Han Li, Theodoros Rekatsinas, AnHai Doan, Youngchoon Park, Ganesh Krishnan, Rohit Deep, Esteban Arcaute, Vijay Raghavendra. 

Name :
Aditya Bhat

Motivation :
Entity matching is used to find the data instances which map to the same real-world entity. This paper explores various state-of-the-art deep learning techniques for this task with different representational power.

Pros :

1. The authors also provide custom representational power based on the choices at each stage of the matching task. The simpler models (SIF) require less time to train whereas the most representational ones (Hybrid) require longer hours to train while providing better predictive power. The medium-complexity models (RNN, Attention) fall in between these two classes of models.

2. The deep learning (DL) techniques proposed in this paper are used for structured, textual and dirty data. This robustness in handling data is very useful when considering larger corpora of data obtained from different sources. The models work better with large training datasets and show significant improvement in performance.

3. The paper proposes a generic framework which can be used to streamline and leverage different DL techniques for the entity matching task. The three steps proposed are Attribute Embedding, Attribute Similarity Representation and Classification. Each and every step can be decoupled, making it easier to plug and play a different type of DL technique which gives us better accuracy. [1,2]

Cons :

1. The DL models require a lot of computation power and also takes a lot of time to train when compared to the Magellan project. The models showed poor performance on clean, structured data when compared against rule-based solutions. This is because in some of the structured datasets the models overfit the data and hence, its performance was affected whereas Magellan was able to perform better because of the restricted search space (during similarity check).

2. The paper does not consider the blocking stage of entity matching, which can be leveraged to produce better pairs of entity mentions. This further enhances the learning phase for the matcher to match the entities and generate lesser false negatives.

3. The paper assumes that the data available is high fidelity labelled data which is generally not the case and a lot of manual effort goes into providing the same. Data in the wild is really dirty and this assumption can hinder the performance of the various models.

Comments

  1. jon_weissmanFebruary 7, 2019 at 7:19 AM

    Good point about the burden of labelling data. The consideration of noisy data with outliers, etc, seems to be a challenging future direction for all of these methods.

    ReplyDelete

Post a Comment

Popular posts from this blog

A Machine Learning Approach to Live Migration Modeling

Dynamic management of resources in data centers improves data center utilization in terms of energy consumption and device utilization. Virtual machines are among techniques to achieve this goal. To enable the VM to adapt to dynamic fluctuations of workloads, Live Migrated virtual machines are introduced, which stands for moving a VM from one physical host to another while the VM keeps running. Live migration enables load balancing, fault tolerance, ease maintenance, etc. However, live migration is costly in terms of data traffic. Having an understanding of performance metrics of different live migration algorithms is important so that one can rely on an efficient algrothm to migrate a running VM is important. Therefore, this work uses a Machine Learning (ML) approach to predict key metrics of different live migration algorithms, and then find the best suitable choice for live migration. Indeed, given resource usage as well as the characteristics of VM work load, this paper predict...
Read more

StormDroid: A Streaminglized Machine Learning-Based System for Detecting Android Malware

Introduction: The paper approaches the problem of Malware Detection using machine learning approach. In this paper, the authors propose considering new type of features along with two traditionally used set of features for developing techniques for malware detection. Along with this proposal, the paper makes other contributions as well. The authors provide a new framework which is able to process information from large data sets as streams, and efficiently perform malware detection. Motivation: Android is an open platform for development. It has certainly made easy for everyone to develop and publish apps.  The number of apps published on Google store is in millions. However, it has also made easier for hackers to reach out to a larger population for malicious attacks. Additionally, with surge in the number of attacks on devices, more data is available for analysis which can be used by the machine learning models to identify malicious apps. This is motivation to use machine ...
Read more

CrystalBall: Statically Analyzing Runtime Behavior via Deep Sequence Learning

Image
Author:  Stephen Zekany, Daniel Rings, Nathan Harada, Michael A. Laurenzano, Lingjia Tang, Jason Mars Intro: Understanding the runtime behavior of the software is critical in many aspects of program development. A conventional approach to this problem is using dynamic profiling, meaning we have to run the program multiple times with environments that representative enough for the actual working situation. Besides that, a dynamic profiling approach has other weakness such as cannot profiling a subset of functions or paths very economically and requires re-profiling each time any changes is applied to the code. This paper presents a statically profiling method, harnessing the ability of RNN to discover the hidden information of hot path in the code. Also, since the training process uses intermediate representation as input, the result will have better generalization power. Methodology: The dataset is in the form of a basic block feature; each sequence of basic blocks repr...
Read more