Masahiro Kajiura

Experiments in Japanese text retrieval and routing using the NEAT system

This paper describes a structured investigation into the retrieval of Japanese text. The study includes a comparison of different indexing strategies for documents and queries, investigation of term weighting strategies principally derived for use with English texts, and the application of relevance feedback for query expansion. Results on the standard BMIR-Jl and BMIR-J2 Japanese retrieval collections indicate that term weighting transfers well to Japanese text. Indexing using dictionary based morphological analysis and character strings are both shown to be individually effective, but marginally better in combination. We also demonstrate that relevance feedback can be used effectively for query expansion in Japanese routing applications.

A first step towards flexible local feedback for ad hoc retrieval

Local feedback for ad hoc retrieval typically hurts performance for about one-third of the search requests while improving the average performance. Our objective is to make it more reliable by estimating the optimal number of assumed-relevant documents and the optimal number of expansion terms for each request. We examine some simple optimization methods based on: the number of case particles in the request; the number of initial search terms; the highest document score in the initial ranked output; and the document score curves. Unfortunately, our first results using the BMIR-J2 and IREX Japanese test collections are negative. We are currently exploring some modified strategies for solving the optimization problems.

Information filtering system operated for NewsWatch’s Daily News Filtering Service

情報通信技術やインフラの向上により日々生成され取得できる情報が増大している中で，情報検索だけではなく必要な情報のみを抽出して提示する情報フィルタリングの必要性が高まっている。株式会社ニューズウォッチは，1996年から新聞を中心とした100を越える媒体からユーザの要求に応じた記事を抽出しユーザに提示する情報フィルタリングサービスを提供している。本記事では，ニューズウォッチ社の各種サービスを支える，自然言語処理技術とWeb技術を用いた情報フィルタリングシステムおよびサービス提供システムについて解説する。

Incremental Relevance Feedback in Japanese Text Retrieval

The application of relevance feedback techniques has been shown to improve retrieval performance for a number of information retrieval tasks. This paper explores incremental relevance feedback for ad hoc Japanese text retrieval; examining, separately and in combination, the utility of term reweighting and query expansion using a probabilistic retrieval model. Retrieval performance is evaluated in terms of standard precision-recall measures, and also using “number-to-view” graphs. Experimental results, on the standard BMIR-J2 Japanese language retrieval collection, show that both term reweighting and query expansion improve retrieval performance. This is reflected in improvements in both precision and recall, but also a reduction in the average number of documents which must be viewed to find a selected number of relevant items. In particular, using a simple simulation of user searching, incremental application of relevance information is shown to lead to progressively improved retrieval performance and an overall reduction in the number of documents that a user must view to find relevant ones.

Recurrent SOLAR algorithm

This paper proposes the Recurrent SOLAR (Supervised One-shot Learning Algorithm for Real number inputs) algorithm, that can complete learning by a single presentation of analog time series data. The most remarkable feature of the Recurrent SOLAR is the one-shot learning, which has been difficult in the past modes, and which can be completed with a high speed by a single presentation of the time-series input data composed of real numbers. The basic idea of the Recurrent SOLAR is the same as in the context model. The time is considered as discrete, and the connection weight of the feedback loop from the output unit or the hidden unit to the state units is fixed. Consequently, the recurrent network can be considered as a feed-forward network. The backpropagation algorithm is used in the context model, while the SOLAR algorithm is applied to the Recurrent SOLAR, which is the one-shot learning algorithm based on the pattern recognition theory. By the use of the SOLAR algorithm, the Recurrent SOLAR can learn the analog time-series data by a single presentation without falling in a local minimum even for large-scale data. In other words, it is a learning algorithm that is suited to an environment where high speed is required in the learning.

Supervised one-shot learning algorithm for real-valued inputs

This paper presents a neural network algorithm called SOLAR (Supervised One-shot Learning Algorithm for Real-valued inputs) that can complete learning by a single presentation of the real-valued inputs. The similarity matrix is introduced in SOLAR, which can measure the Euclidean distance between training instances represented by a real number. Based on the geometrical structure of the similarity matrix, the structure of the network, the connection weights, the learning parameters and the linear discriminant function are determined. Then, the learning is executed by a single presentation of the training set. The nonlinearly separable case can be reduced to the linearly separable case by decomposing the set of supervisor signals into linearly separable sets. The main contribution of this paper is that SOLAR realizes a high-speed learning by a single presentation of the input data composed of real values and a large-scale complex set of data can be learned. This has been difficult in previous models. In addition, since the number of hidden units and the parameters are determined by the algorithm, there is no problem such as trial-and error. Simulations show the effectiveness of the proposed method by learning a relatively large-scale problem, such as the backpropagation algorithm, which has been difficult based on the gradient descent method.