EAGER: Supporting GUI-Based Text Analytics on Social Media Data by Non-Technical Users

大量信息正以越来越快的速度从Twitter等社交媒体来源中被创造出来。许多研究人员需要对社交媒体数据进行文本分析，以获得特定领域的见解，通过执行各种计算任务，如关键字搜索、正则表达式、自然语言处理和情感分析。一个主要的挑战是他们缺乏IT背景，这使得他们很难有效地进行研究。此外，他们经常需要使用机器学习模型来进行深度分析，而为训练模型生成标记实例的数据准备过程可能是耗时和劳动密集型的。该项目研究如何支持没有强大IT背景的用户对社交媒体数据进行文本分析。它开发了一个开源系统，其目标如下：(1)将常见的文本计算作为基本运算符进行模块化；(2)提供图形用户界面（GUI），用户无需编写代码即可声明式地形成工作流计划；(3)支持文本分析作为基于web的服务；(4)尽可能支持索引以提高性能。该系统已经在Github上开发了超过15个月，有30多名提交者和21000行高质量的源代码。加州大学欧文分校（UC Irvine）的公共卫生研究人员建立了一个最初的原型，并使用它来分析Twitter数据。这些技术是通用的、面向系统的，也可以使许多其他领域受益。

A wealth of information is being created at an increasingly fast rate from social media sources such as Twitter. Many researchers need text analytics on social media data to obtain domain-specific insights, by doing various computational tasks such as keyword search, regular expression, natural language processing, and sentiment analysis. A main challenge is their lack of IT background, making it hard for them to conduct research efficiently. In addition, very often they need to use machine learning models to do deep analysis, and the data-preparation process for generating labeled instances for training models can be time consuming and labor intensive. This project studies how to support text analytics on social media data by users who do not have a strong IT background. It develops an open-source system with the following goals: (1) modularizing common text computation as basic operators; (2) providing a graphic user interface (GUI) for users to form a workflow plan declaratively without writing code; (3) supporting text analytics as a Web-based service; and (4) supporting indexing whenever possible to improve performance. The system has been under development on Github for more than 15 months with more than 30 committers and 21 thousand lines of high-quality source code. An initial prototype is setup and used by Public Health researchers at UC Irvine to analyze Twitter data. The techniques are general-purpose, systems-oriented, and can benefit many other domains as well.