Heeryon Cho

Data-driven feature word selection for clustering online news comments

Popular news articles attract thousands of online comments, making it tedious and time-consuming for a manual review. Automatically clustering similar comments can help reduce the burden of manual analyses, but appropriate feature words must be selected for successful clustering. In this paper, we present a data-driven feature word selection method which realizes structurally superior clustering of online comments. The top 1,000 most frequent nouns appearing across the entire 7.44 million Korean online comments are selected to construct an overall noun set. Frequent nouns in the online comments of each news article are selected to construct the local noun set. The intersection between the local and overall noun set is taken to construct the global noun set. The global noun set is removed from the corresponding local noun set to construct the distinct noun set. The top 250 most frequent nouns are selected for each of the local, global, and distinct noun sets for K-means clustering. The clustered results are evaluated using three internal cluster validation indices, Dunn, PBM, and Silhouette. As a result, online comments clustered using distinct nouns produced structurally superior clusters when compared to the other types of nouns, local and global.

Structure preserving dimensionality reduction for visual object recognition

Robust object recognition has drawn increasing attention in the field of computer vision and machine learning with fast development in feature extraction and classification techniques, and release of public datasets, such as Caltech datasets, Pascal Visual Object Classes, and ImageNet. Recently, deep learning based object recognition systems have shown significant performance improvements in visual object recognition tasks using innovative learning methodology. However, high dimensional space searching and recognition is time consuming, so performing point and range queries in high dimension is reconsidered for object recognition. This paper proposes optimized dimensionality reduction using structured sparse principle component analysis. The proposed method retains high dimensional feature structures, removes redundant features that do not contribute to similarity, and classifies the query image in a large database. The qualitative and quantitative experimental results, including a comparison with the current state-of-the-art visual object recognition algorithms, verify that the proposed recognition algorithm performs favorably in reducing the query image dimension and number of training images.

Statistical noise reduction for robust human activity recognition

Noise and variability in accelerometer data collected using smart devices obscure accurate human activity recognition. In order to tackle the degradation of the triaxial accelerometer data caused by noise and individual user differences, we propose a statistical noise reduction method using total variation minimization to attenuate the noise mixed in the magnitude feature vector generated from triaxial accelerometer data. The experimental results using Random Forest classifier prove that our noise removal approach is constructive in significantly improving the human activity recognition performance.

Improving sentiment classification through distinct word selection

While the performance of sentiment classification has steadily risen through the introduction of various feature-based methods and distributed representation-based approaches, less attention was given to the qualitative aspect of classification, for instance, the identification of useful words in individual opinion texts. We present an approach using set operations for identifying useful words for sentiment classification, and employ truncated singular value decomposition (SVD), a classic low-rank matrix decomposition technique for document retrieval, in order to tackle the issue of both synonymy and noise removal. The sentiment classification performance of our approach, which concatenates three kinds of features, outperforms the existing word-based and distributed word representation-based methods and is comparable to the existing state of the art distributed document representation-based approaches.

Target Object Tracking-Based 3D Object Reconstruction in a Multiple Camera Environment in Real Time

The visualization of a three-dimensional target object reconstruction from multiple cameras is an important issue in high-dimensional data representations with application for medical uses, sports scene analysis, and event creation for film. In this paper, we propose an efficient 3D reconstruction methodology to voxelize and carve the 3D scene in focus on 3D tracking of the object in a large environment. We applied sparse representation-based target object tracking to efficiently trace the movement of the target object in a background clutter and reconstruct the object based on the estimated 3D position captured from multiple images. The voxelized area is optimized to the target by tracking the 3D position and then effectively reduce the process time while keeping the details of the target. We demonstrate the experiments by carving the voxels within the 3D tracked area of the target object.

Feature word selection by iterative top-K aggregation for classifying recommended shops

We propose a feature word selection method for classifying recommended shops using Yelp customer reviews. TextRank keywords are extracted from the customer reviews to construct the sorted positive and negative keyword lists based on each keywords summed TextRank scores. The top-K keywords are then aggregated iteratively by multiples of K to construct the positive and negative keyword frequency lists. The negative keyword frequency list is then subtracted from the positive keyword frequency list, and the resulting list is standardized to generate the final positive and negative keyword lists. The performance of our feature selection method is evaluated using Naïve Bayes classifiers, and the binary classification accuracy of the selected feature words is 77.94%, which is better than the baseline χ2 feature word selection.

Retrieving Minority Product Reviews Using Positive/Negative Skewness

A given product’s online product reviews build up to form largely positive or negative reviews or mixed reviews that include both the positive and negative reviews. While the homogeneously positive or negative reviews help readers identify the generally praised or criticized product, the mixed reviews with minority opinions potentially contain valuable information about the product. We present a method of retrieving minority opinions from the online product reviews using the skewness of positive/negative reviews. The proposed method first classifies the positive/negative product reviews using a sentiment dictionary and then calculates the skewness of the classified results to identify minority reviews. Minority review retrieval experiments were conducted on smartphone and movie reviews, and the F1-measures were 24.6% (smartphone) and 15.9% (movie) and the accuracies were 56.8% and 46.8% when the individual reviews’ sentiment classification accuracies were 85.3% and 78.8%. The theoretical performance of minority review retrieval is also discussed.

Automatic and Dynamic Book Category Assignment Using Concept-Based Book Ontology

We propose concept-based book ontology for automatic and dynamic category assignment to books through collaborative filtering. It is general for authors or book systems to assign one or more categories to books, but determining book categories based on book reviews have long been neglected. Popularization of online reviews has generated abundant reviews, and it is valuable to additively consider these reviews for assigning relevant book categories. The proposed concept-based book ontology is constructed by conceptual categories that are extracted from the existing book category hierarchy using semantic relationships. Moreover, category-specific review words are constructed through collaborative filtering with the semantically related review words. We built an automatic and dynamic book category assignment prototype system using the concept-based book ontology with the Amazon book department data and confirmed the effectiveness of our approach through empirical evaluations.