Rey-Long Liu

Incremental context mining for adaptive document classification

Automatic document classification (DC) is essential for the management of information and knowledge. This paper explores two practical issues in DC: (1) each document has its context of discussion, and (2) both the content and vocabulary of the document database is intrinsically evolving. The issues call for adaptive document classification (ADC) that adapts a DC system to the evolving contextual requirement of each document category, so that input documents may be classified based on their contexts of discussion. We present an incremental context mining technique to tackle the challenges of ADC. Theoretical analyses and empirical results show that, given a text hierarchy, the mining technique is efficient in incrementally maintaining the evolving contextual requirement of each category. Based on the contextual requirements mined by the system, higher-precision DC may be achieved with better efficiency.

AN EMPIRICAL STUD' ON THEMATIC KNOWLEDGE ACQUISITION BASED ON SYNTACTIC CLUES AND HEURISTICS

Thematic knowledge is a basis of semantic interpretation. In this paper, we propose an acquisition method to acquire thematic knowledge by exploiting syntactic clues from training sentences. The syntactic clues, which may be easily collected by most existing syntactic processors, reduce the hypothesis space of the thematic roles. The ambiguities may be further resolved by the evidences either from a trainer or from a large corpus. A set of heuristics based on linguistic constraints is employed to guide the ambiguity resolution process. When a trainer is available, the system generates new sentences whose thematic validities can be justified by the trainer. When a large corpus is available, the thematic validity may be justified by observing the sentences in the corpus. Using this way, a syntactic processor may become a thematic recognizer by simply deriving its thematic knowledge from its own syntactic knowledge.

Adaptive exception monitoring agents for management by exceptions

Management by exceptions (MBE) is an effective management strategy in many domains. It suggests that managers focus on important jobs (e.g., planning and decision-making) without being involved in the tedious monitoring of exceptions (e.g., a critical item whose current status violates some regulations). Once an exception is detected, the managers are notified to respond to the exception promptly. Therefore, exception monitoring is the key to realize the idea of MBE. An exception monitoring system should detect exceptions in a timely manner for the managers. It should also control the extra loading it incurs to related information servers (e.g., database management systems) and the Intranet, which are fundamental backbones for information processing in businesses. In this paper, a multiagent paradigm Adaptive Agents for Management by Exceptions (AAMBE) is proposed for exception monitoring. The agents adapt to the environment by learning to work together to achieve timely detection of exceptions. An experiment to investigate the performance of AAMBE is conducted by simulating real-world operations of financial management in merchandising trades. Empirical and theoretical analyses show that AAMBE may detect more exceptions in a timelier manner by incurring less extra loading to the related information servers and the Intranet.

Adaptive sampling for thresholding in document filtering and classification

Document filtering (DF) and document classification (DC) are often integrated together to classify suitable documents into suitable categories. A popular way to achieve integrated DF and DC is to associate each category with a threshold. A document d may be classified into a category c only if its degree of acceptance (DOA) with respect to c is higher than the threshold of c. Therefore, tuning a proper threshold for each category is essential. A threshold that is too high (low) may mislead the classifier to reject (accept) too many documents. Unfortunately, thresholding is often based on the classifiers DOA estimations, which cannot always be reliable, due to two common phenomena: (1) the DOA estimations made by the classifier cannot always be correct, and (2) not all documents may be classified without any controversy. Unreliable estimations are actually noises that may mislead the thresholding process. In this paper, we present an adaptive and parameter-free technique AS4T to sample reliable DOA estimations for thresholding. AS4T operates by adapting to the classifiers status, without needing to define any parameters. Experimental results show that, by helping to derive more proper thresholds, AS4T may guide various classifiers to achieve significantly better and more stable performances under different circumstances. The contributions are of practical significance for real-world integrated DF and DC.

Interactive high-quality text classification

Automatic text classification (TC) is essential for information sharing and management. Its ideal goals are to achieve high-quality TC: (1) accepting almost all documents that should be accepted (i.e., high recall) and (2) rejecting almost all documents that should be rejected (i.e., high precision). Unfortunately, the ideal goals are rarely achieved, making automatic TC not suitable for those applications in which a classifiers erroneous decision may incur high cost and/or serious problems. One way to pursue the ideal is to consult users to confirm the classifiers decisions so that potential errors may be corrected. However, its main challenge lies on the control of the number of confirmations, which may incur heavy cognitive load on the users. We thus develop an intelligent and classifier-independent confirmation strategy ICCOM. Empirical evaluation shows that ICCOM may help various kinds of classifiers to achieve very high precision and recall by conducting fewer confirmations. The contributions are significant to the archiving and recommendation of critical information, since identification of possible TC errors (those that require confirmation) is the key to process information more properly.

Incremental mining of information interest for personalized web scanning

Businesses and people often organize their information of interest (IOI) into a hierarchy of folders (or categories). The personalized folder hierarchy provides a natural way for each of the users to manage and utilize his/her IOI (a folder corresponds to an interest type). Since the interest is relatively long-term, continuous web scanning is essential. It should be directed by precise and comprehensible specifications of the interest. A precise specification may direct the scanner to those spaces that deserve scanning, while a specification comprehensible to the user may facilitate manual refinement, and a specification comprehensible to information providers (e.g. Internet search engines) may facilitate the identification of proper seed sites to start scanning. However, expressing such specifications is quite difficult (and even implausible) for the user, since each interest type is often implicitly and collectively defined by the content (i.e. documents) of the corresponding folder, which may even evolve over time. In this paper, we present an incremental text mining technique to efficiently identify the users current interest by mining the users information folders. The specification mined for each interest type specifies the context of the interest type in conjunctive normal form, which is comprehensible to general users and information providers. The specification is also shown to be more precise in directing the scanner to those sites that are more likely to provide IOI. The user may thus maintain his/her folders and then constantly get IOI, without paying much attention to the difficult tasks of interest specification and seed identification.

Ranker enhancement for proximity-based ranking of biomedical texts

Biomedical decision making often requires relevant evidence from the biomedical literature. Retrieval of the evidence calls for a system that receives a natural language query for a biomedical information need and, among the huge amount of texts retrieved for the query, ranks relevant texts higher for further processing. However, state-of-the-art text rankers have weaknesses in dealing with biomedical queries, which often consist of several correlating concepts and prefer those texts that completely talk about the concepts. In this article, we present a technique, Proximity-Based Ranker Enhancer (PRE), to enhance text rankers by term-proximity information. PRE assesses the term frequency (TF) of each term in the text by integrating three types of term proximity to measure the contextual completeness of query terms appearing in nearby areas in the text being ranked. Therefore, PRE may serve as a preprocessor for (or supplement to) those rankers that consider TF in ranking, without the need to change the algorithms and development processes of the rankers. Empirical evaluation shows that PRE significantly improves various kinds of text rankers, and when compared with several state-of-the-art techniques that enhance rankers by term-proximity information, PRE may more stably and significantly enhance the rankers.

Text classification for healthcare information support

Healthcare information support (HIS) is essential in managing, gathering, and disseminating information for healthcare decision support through the Internet. To support HIS, text classification (TC) is a key kernel. Upon receiving a text of healthcare need (e.g. symptom description from patients) or healthcare information (e.g. information from medical literature and news), a text classifier may determine its corresponding categories (e.g. diseases), and hence subsequent HIS tasks (e.g. online healthcare consultancy and information recommendation) may be conducted. The key challenge lies on high-quality TC, which aims to classify most texts into suitable categories (i.e. recall is very high), while at the same time, avoid misclassifications of most texts (precision is very high). High-quality TC is particularly essential, since healthcare is a domain where an error may incur higher cost and/or serious problems. Unfortunately, high-quality TC was seldom achieved in previous studies. In the paper, we present a case study in which a high-quality classifier is built to support HIS in Chinese disease-related information, including the cause, symptom, curing, side-effect, and prevention of cancer. The results show that, without relying on domain knowledge and complicated processing, cancer information may be classified into suitable categories, with a controlled amount of confirmations.

Context-based online medical terminology navigation

The World Wide Web is an important hyperspace for health information navigation. To comprehend health information, medical terminology is a main obstacle. Therefore, starting from a medical term t in a document d being read, information consumers often need to navigate to those documents that are relevant to t under the current context of interest (COI), which may be indicated by the background content of d (e.g., health topic discussed in d). However, health information consumers are often unable or unwilling to construct a proper query for the navigation. Therefore, we present a technique COMTN to automatically generate a context-indicative query for a term designated by an information consumer when reading a document. In addition to the term, the query is enhanced with those terms (from the document) that may indicate the current COI of the consumer. Empirical results show that, using the queries generated by COMTN, more relevant information about the designated term may be retrieved from MedlinePlus. Moreover, COMTN is efficient in online query generation, and it may work for evolving health terminology without training. The contributions are significant to online medical knowledge discovery and comprehension, which are essential factors to promote the value of health information.

Collaborative Multiagent Adaptation for Business Environmental Scanning Through the Internet

Businesses should promptly respond to their dynamic environments. Environmental scanners are thus essential for the businesses to discover and monitor environmental information of interest (IOI). In this paper, we explore user-centered, continuous and resource-bounded environmental scanning (UCRES). Upon receiving information preferences of managers, new IOI should be continuously detected in a timely and complete manner without consuming too much resource (e.g. bandwidths of computer networks and services of information servers). We develop a multiagent framework AESA to tackle the challenges of UCRES. Each agent is a simple entity. All agents collaboratively adapt their population and resource consumption to several dynamic aspects of UCRES: information preferences of individual users, resource limitation of environmental scanning, distribution of IOI in the environments, and update behaviors of the IOI. The delivery of AESA to businesses may constantly provide a larger amount of important and timely IOI without exhausting the Intranet and the Internet communities.