Influence of Graph Construction on Semi-supervised Learning
نویسندگان
چکیده
A variety of graph-based semi-supervised learning (SSL) algorithms and graph construction methods have been proposed in the last few years. Despite their apparent empirical success, the field of SSL lacks a detailed study that empirically evaluates the influence of graph construction on SSL. In this paper we provide such an experimental study. We combine a variety of graph construction methods as well as a variety of graph-based SSL algorithms and empirically compare them on a number of benchmark data sets widely used in the SSL literature. The empirical evaluation proposed in this paper is subdivided into four parts: (1) best case analysis; (2) classifiers’ stability evaluation; (3) influence of graph construction; and (4) influence of regularization parameters. The purpose of our experiments is to evaluate the trade-off between classification performance and stability of the SSL algorithms on a variety of graph construction methods and parameter values. The obtained results show that the mutual k-nearest neighbors (mutKNN) graph may be the best choice for adjacency graph construction while the RBF kernel may be the best choice for weighted matrix generation. In addition, mutKNN tends to generate smoother error surfaces than other adjacency graph construction methods. However, mutKNN is unstable for a relatively small value of k. Our results indicate that the classification performance of the graph-based SSL algorithms are heavily influenced by the parameters setting and we found no evident explorable pattern to relay to future practitioners. We discuss the consequences of such instability in research and practice.
منابع مشابه
Network Construction and Applications for Semi-Supervised Learning
The influence of network construction on graphbased semi-supervised learning (SSL) and their related applications have only received limited study despite its critical impact on accuracy. We introduce four variants for network construction for SSL that adopt different network topology: 1) S-kNN (Sequential k-Nearest Neighbors) that generates regular networks; 2) GBILI (Graph Based on the inform...
متن کاملTopics in Graph Construction for Semi-Supervised Learning
Graph-based Semi-Supervised Learning (SSL) methods have had empirical success in a variety of domains, ranging from natural language processing to bioinformatics. Such methods consist of two phases. In the first phase, a graph is constructed from the available data; in the second phase labels are inferred for unlabeled nodes in the constructed graph. While many algorithms have been developed fo...
متن کاملA Graph-Based Semi-Supervised Learning for Question Semantic Labeling
We investigate a graph-based semi-supervised learning approach for labeling semantic components of questions such as topic, focus, event, etc., for question understanding task. We focus on graph construction to handle learning with dense/sparse graphs and present Relaxed Linear Neighborhoods method, in which each node is linearly constructed from varying sizes of its neighbors based on the dens...
متن کاملData-Driven Graph Construction for Semi-Supervised Graph-Based Learning in NLP
Graph-based semi-supervised learning has recently emerged as a promising approach to data-sparse learning problems in natural language processing. All graph-based algorithms rely on a graph that jointly represents labeled and unlabeled data points. The problem of how to best construct this graph remains largely unsolved. In this paper we introduce a data-driven method that optimizes the represe...
متن کاملSupervised neighborhood graph construction for semi-supervised classification
Graph based methods are among the most active and applicable approaches studied in semi-supervised learning. The problem of neighborhood graph construction for these methods is addressed in this paper. Neighborhood graph construction plays a key role in the quality of the classification in graph based methods. Several unsupervised graph construction methods have been proposed that have addresse...
متن کامل