VLDB-2019/8-TextCube: Automated Construction and Multidimensional Exploration

[BrambleXu]

Summary:

提出了一个TextCube的数据结构框架。讲了为了做到自动化构建这个框架，用到了哪些技术。#275 的团队。

Resource:

• pdf
• [code](
• [paper-with-code](

Paper information:

• Author:
• Dataset:
• keywords:

Notes:

TextCube provides a critical information organization structure, enhancing text
exploration and analysis for various applications.

We focus on new TextCube construction methods that are scalable, weakly-supervised, domain-independent, language-agnostic, and effective (i.e., generating quality TextCubes from large corpora of various domains).

Module I. Mining Structural Primitives from Text: Phrases, Entities and Relations

• AutoPhrase
• AutoNER
• ReMine [58] which extracts high-confidence relational phrases from domain-specific texts in an end-to-end manner.

Module II. Automated Construction of TextCubes

1. Taxonomy construction: Taxonomy construction clusters similar concepts and generates a hierarchy of “concept clusters” from massive corpus。 模型：TaxoGen [53], a recursive framework that leverages word distributional representations and constructs cluster-based taxonomy using adaptive spherical clustering and local embedding

2. Embedding learning: serve as the preliminary to document classification and TextCube construction. 模型：JoSE. , an unsupervised text embedding framework that jointly learns word embedding and paragraph embedding by incorporating both local and global contexts to capture more complete text semantics, and present TopicMine [24], a category-name guided word embedding framework that endows word embedding with discriminative power over the specific set of categories

3. Supervised methods: for text cube construction。 We present how to adapt the supervised methods for text cube construction along with their strength and drawbacks.

4. Weakly-supervised methods: WeSTClass [25] and WeSHClass [26], which generate pseudo training data for neural classifier pre-training, and then bootstrap the classifier by selftraining on unlabeled documents.

Module III. Multi-Dimensional Exploration of TextCubes

TextCube facilitates multidimensional text analysis

1. Cube-based multidimensional analysis:
2. Text summarization:

Model Graph:

Result:：

Thoughts:

Next Reading: