Sylvester Olubolu Orimaye

Learning Predictive Linguistic Features for Alzheimer’s Disease and related Dementias using Verbal Utterances

Early diagnosis of neurodegenerative disorders (ND) such as Alzheimer’s disease (AD) and related Dementias is currently a challenge. Currently, AD can only be diagnosed by examining the patient’s brain after death and Dementia is diagnosed typically through consensus using specific diagnostic criteria and extensive neuropsychological examinations with tools such as the Mini-Mental State Examination (MMSE) or the Montreal Cognitive Assessment (MoCA). In this paper, we use several Machine Learning (ML) algorithms to build diagnostic models using syntactic and lexical features resulting from verbal utterances of AD and related Dementia patients. We emphasize that the best diagnostic model distinguished the AD and related Dementias group from the healthy elderly group with 74% FMeasure using Support Vector Machines (SVM). Additionally, we perform several statistical tests to indicate the significance of the selected linguistic features. Our results show that syntactic and lexical features could be good indicative features for helping to diagnose AD and related Dementias.

Performance and trends in recent opinion retrieval techniques

This paper presents trends and performance of opinion retrieval techniques proposed within the last 8 years. We identify major techniques in opinion retrieval and group them into four popular categories. We describe the state-of-the-art techniques for each category and emphasize on their performance and limitations. We then summarize with a performance comparison table for the techniques on different datasets. Finally, we highlight possible future research directions that can help solve existing challenges in opinion retrieval.

Buy it - don't buy it: sentiment classification on amazon reviews using sentence polarity shift

In recent years, sentiment classification has been an appealing task for so many reasons. However, the subtle manner in which people write reviews has made achieving high accuracy more challenging. In this paper, we investigate the improvements on sentiment classification baselines using sentiment polarity shift in reviews. We focus on Amazon online reviews for different types of product. First, we use our newly-proposed Sentence Polarity Shift (SPS) algorithm on review documents, reducing the relative classification loss due to inconsistent sentiment polarities within reviews by an average of 16% over a supervised sentiment classifier. Second, we build up on a popular supervised sentiment classification baseline by adding different features which provide better improvement over the original baseline. The improvement shown by this technique suggests modeling sentiment classification systems based on polarity shift combined with sentence and document-level features.

Predicting probable Alzheimer’s disease using linguistic deficits and biomarkers

BackgroundThe manual diagnosis of neurodegenerative disorders such as Alzheimer’s disease (AD) and related Dementias has been a challenge. Currently, these disorders are diagnosed using specific clinical diagnostic criteria and neuropsychological examinations. The use of several Machine Learning algorithms to build automated diagnostic models using low-level linguistic features resulting from verbal utterances could aid diagnosis of patients with probable AD from a large population. For this purpose, we developed different Machine Learning models on the DementiaBank language transcript clinical dataset, consisting of 99 patients with probable AD and 99 healthy controls.ResultsOur models learned several syntactic, lexical, and n-gram linguistic biomarkers to distinguish the probable AD group from the healthy group. In contrast to the healthy group, we found that the probable AD patients had significantly less usage of syntactic components and significantly higher usage of lexical components in their language. Also, we observed a significant difference in the use of n-grams as the healthy group were able to identify and make sense of more objects in their n-grams than the probable AD group. As such, our best diagnostic model significantly distinguished the probable AD group from the healthy elderly group with a better Area Under the Receiving Operating Characteristics Curve (AUC) using the Support Vector Machines (SVM).ConclusionsExperimental and statistical evaluations suggest that using ML algorithms for learning linguistic biomarkers from the verbal utterances of elderly individuals could help the clinical diagnosis of probable AD. We emphasise that the best ML model for predicting the disease group combines significant syntactic, lexical and top n-gram features. However, there is a need to train the diagnostic models on larger datasets, which could lead to a better AUC and clinical diagnosis of probable AD.

Can predicate-argument structures be used for contextual opinion retrieval from blogs?

We present the results of our investigation on the use of predicate-argument structures for contextual opinion retrieval. The use of predicate-argument structure for opinion retrieval is a novel approach that exploits the grammatical derivation of sentences to show contextual and subjective relevance. We do not use frequency of certain keywords as it is usually done in keyword-based opinion retrieval approaches. Rather, our novel solution is based on frequency of contextually relevant and subjective sentences. We use a linear relevance model that leverages semantic similarities among predicate-argument structures of sentences. Thus, this paper presents the evaluation results of the linear relevance model. The model does a linear combination of a popular relevance model, our proposed transformed terms similarity model, and the absolute value of a sentence subjectivity scoring scheme. The predicate-argument structures are derived from the grammatical derivations of natural language query topics and the well formed sentences from blog documents. The derived predicate-argument structures are then semantically compared to compute an opinion relevance score. Our scoring technique uses the highest frequency of semantically related predicate-argument structures enriched with the total subjectivity score from sentences. Evaluation and experimental results show that predicate-argument structures can indeed be used for contextual opinion retrieval as it improves performance of opinion retrieval task by 15% over the popular TREC baselines.

Learning to Classify Subjective Sentences from Multiple Domains Using Extended Subjectivity Lexicon and Subjective Predicates

We investigate the performance of subjective predicates and other extended predictive features on subjectivity classification in and across different domains. Our approach constructs a semi-supervised subjective classifier based on an extended subjectivity lexicon that includes subjective annotations resulting from a manually annotated subjectivity corpus, a list of manually constructed subjectivity clues, and a set of subjective predicates learned from a large collection of likely subjective sentences. Using the extended lexicon, we extracted high precision subjective sentences from multiple domains and constructed in-domain and cross-domain subjectivity classifiers. Experimental results on multiple datasets show that the proposed technique performed comparatively better than a high precision subjectivity classification baseline and has improved cross-domain accuracy. We report 97.7% precision, 73.4% recall and 83.8% F-Measure for in-domain subjectivity classification and a accuracy level of 84.6% for cross-domain subjectivity classification.

Natural language opinion search on blogs

In this paper, we present natural language opinion search by unifying discourse representation structures and the subjectivity of sentences to search for relevant opinionated documents. This technique differs from existing keyword-based opinion retrieval techniques which do not consider semantic relevance of opinionated documents at discourse level. We propose a simple message model that uses the attributes of the discourse representation structures and a list of opinion words. The model compute the relevance and opinionated scores of each sentence to a given query topic. We show that the message model is able to effectively identify which entity in a sentence is directly affected by the presence of opinion words. Thus, opinionated documents containing relevant topic discourse structures are retrieved based on the instances of opinion words that directly affect the key entities in relevant sentences. In terms of MAP, experimental results show that the technique retrieves opinionated documents with better results than the standard TREC Blog 08 best run, a non-proximity technique, and a state-of-the-art proximity-based technique.

Sentence-level contextual opinion retrieval

Existing opinion retrieval techniques do not provide context-dependent relevant results. Most of the approaches used by state-of-the-art techniques are based on frequency of query terms, such that all documents containing query terms are retrieved, regardless of contextual relevance to the intent of the human seeking the opinion. However, in a particular opinionated document, words could occur in different contexts, yet meet the frequency attached to a certain opinion threshold, thus explicitly creating a bias in overall opinion retrieved. In this paper we propose a sentence-level contextual model for opinion retrieval using grammatical tree derivations and approval voting mechanism. Model evaluation performed between our contextual model, BM25, and language model shows that the model can be effective for contextual opinion retrieval such as faceted opinion retrieval.

Emerging approaches in literature-based discovery: techniques and performance review

Literature-based discovery systems aim at discovering valuable latent connections between previously disparate research areas. This is achieved by analyzing the contents of their respective literatures with the help of various intelligent computational techniques. In this paper, we review the progress of literature-based discovery research, focusing on understanding their technical features and evaluating their performance. The present literature-based discovery techniques can be divided into two general approaches: the traditional approach and the emerging approach. The traditional approach, which dominate the current research landscape, comprises mainly of techniques that rely on utilizing lexical statistics, knowledge-based and visualization methods in order to address literature-based discovery problems. On the other hand, we have also observed the births of new trends and unprecedented paradigm shifts among the recently emerging literature-based discovery approach. These trends are likely to shape the future trajectory of the next generation literature-based discovery systems.