Amyotrophic lateral sclerosis (ALS) is a serious disease that affects nerve and muscle function, with no known cure. Early and accurate monitoring is essen tial to help physicians provide better care. Although machine learning has been applied to predict the progression of ALS, many models struggle with issues such as poor data quality and missing information, which affect accuracy. In this paper, our aim is to improve existing models by introducing better features to enhance prediction performance. A key contribution is the development of a new feature called the physical health index (PHI), which combines four im portant patient attributes: body mass index (BMI), weight, forced vital capacity (FVC), and basal calories. This feature provides a clearer view of the physical health of the patient, enabling the model to learn more effectively. We used the IDPP CLEF 2024 BTO dataset and performed three experiments: using 50 raw features, 29 engineered features, and 25 further engineered features including PHI. The results showed that the R-squared of the XGBoost model improved from 0.9573 to 0.9663 and finally 0.9828, while RMSE decreased from 0.2317 to 0.1801 and then 0.1182 with PHI. This study highlights how targeted feature engineering can improve the prediction of ALS using machine learning.

In this study we investigate different strategies for enriching tweets (now posts) to improve tweets classification accuracy. In particular, we investigate the effects of emoji-to-text translation and semantic enrichment using a locally trained word2vec model. Before doing the enrichment, we develop a baseline system where we investigate the effects of stopwords removal. In our investigation, we found out that stopwords removal degrades the classification accuracy of tweets. In the first approach, we compare the difference in the classification accuracies between tweets with stopwords and tweets without stopwords. In the second approach, we enrich tweets by translating emojis contained in tweets to their equivalent text. To test the accuracy of our approaches, we employed 3 classifiers, namely, BERT classifier, Random Forest and Multinomial Naive Bayes. The third approach builds on the results of the first approach, and it uses a locally trained word2vec model to expand tweets. The results of our study suggest that semantic enrichment and translating emojis to their equivalent meaning increases the classification accuracy of tweets.

In this paper, an investigation was done to identify writing style features that can be used for cross-topic and cross-genre documents in the Authorship Identification task from 2003 to 2015. Different writing style features were empirically evaluated that were previously used in single topic and single genre documents for Authorship Identification to determine whether they can be used effectively for cross-topic and crossgenre Authorship Identification using an ablation process. The dataset used was taken from the 2015 PAN CLEF Forum English collection consisting of 100 sets. Furthermore, it was investigated whether combining some of these feature sets can help improve the authorship identification task. Three different classifiers were used: Naïve Bayes, Support Vector Machine, and Random Forest. The results suggest that a combination of a lexical, syntactical, structural, and content feature set can be used effectively for cross topic and cross genre authorship identification, as it achieved an AUC result of 0.837.

In this paper, team University of Botswana Computer Science (UBCS) investigate the opinions of Twitter users towards vaccine uptake. In particular, we build three different text classifiers to detect people's opinions and classify them as provax-for opinions that are for vaccination, antivax for opinions against vaccination and neutral-for opinions that are neither for or against vaccination. Two different datasets obtained from Twitter, 1 by Cotfas and the other by Fire2022 Organizing team were merged to and used for this study. The dataset contained 4392 tweets. Our first classifier was based on the basic BERT model and the other 2 were machine learning models, Random Forest and Multinomial Naive Bayes models. Naive Bayes classifier outperformed other classifiers with a macro-F1 score of 0.319. © 2022 Copyright for this paper by its authors.

In this paper, team University of Botswana Computer Science (UB-CS) investigate the use of XLM-RoBERTa, a multilingualmodeltrainedon100 different languagesfortransfer learninginthe identification of conversational hate-speech in code-mixed languages. We also investigate whether enriching the tweets with textual sentiments from emojis can help improve the classification performance. Our proposed solution outperformed other teams that participated at the HASOC (2022) Task 2 with a macro F1 score of 0.4939. The result suggest that enriching the tweets with textual sentiments and using a pre-trained multilingual model for transfer learning can help in the identification of conversational hate-speech in code-mixed languages.

Road traffic accidents are a serious problem for the nation of Botswana. A large amount of money is used to compensate those who are affected by road accidents. Traffic accidents are one of the major causes of Deaths in Botswana. It is important for relevant organizations to have a reliable source of data for accurate evaluation of traffic accidents. Similarly, data on vehicle registration must be transformed and be readily available to assist managerial decision makers. In this article, we deploy a Business Intelligence (BI) and Data Warehouse (DW) solution in an attempt to assist the relevant departments in their road traffic accidents and vehicle registration evaluation. In Our evaluation of the traffic accidents our findings suggest that across accident severity, Damage Only accidents had the most interesting recent trend with a 11.93% decrease in the last 3 years on record. Count of Accident Severity for Damage Only accidents dropped from 13,491 to 11,881 between 2018 and 2020 whilst Minor accidents experienced the longest period of growth. Most accidents take place in rural locations and more accidents take place during the weekend. At 28,439, Sunday had the highest number of accidents and was 47.59% higher than Wednesday, which had the lowest count of accidents at 19,269. The results for vehicle registration reveal that the number of vehicle registration decreased for the last 3 years on record. The number of vehicles registered dropped from 65535 to 24457 during its steepest decline between 2019 and 2021.

In this paper, team University of Botswana Computer Science (UBCS) investigate whether enriching social media data with text generated from emojis can help in the identification of Hate Speech and Offensive Content. In particular, we build three different binary text classifiers that can detect Hate and Offensive content (HOF) or Not Hate-Offensive content (NOT) on data sampled from Twitter. In building our first classifier, we used pre-processed text from twitter only without emojis. In the second classifier, we enrich our preprocessed text from Twitter with text generated from emojis within the Tweets. Our result suggests that enriching Tweets with text generated from emojis within the Tweets improves the classification accuracy of our hate and offensive content classier.

This paper describes our participating systems in the FIRE AILA 2021 shared task on predicting rhetorical roles for sentences in a legal judgement document. In particular we propose three multi-class classifiers to predict for each of the sentences a rhetorical role from the following: facts, arguments, ratio of the decision, precedent, statutes, ruling of lower court and ruling of present court. Each of the classifiers uses a supervised fastText model. As input tokens the first classifier uses unigrams, the second one used bigrams and the last one uses trigrams. Our system that uses trigrams attains an F-Score of 0.340 followed closely by the bigram system at 0.338 while the baseline has a score of 0.317.

. In this paper, we explore three different ad hoc retrieval approaches to rank verified claims, so that those that verify the input claim are ranked on top. In particular, we deploy DPH Divergence from Ran-domness (DFR) term weighting model to rank the verified claims. In addition, we deploy the Sequential Dependence (SD) variant of the Markov Random Fields (MRF) for term dependence to re-rank documents (veri-fied claims) that have query terms (input claim) in close proximity. More-over, we deploy LambdaMART, which is a learning to rank algorithm that use machine learning techniques to learn an appropriate combination of features into an effective ranking model.

In this paper we explore several retrieval strategies in an attempt to identify relevant statues and prior cases using a description of a current situation (current case). In particular, we investigate whether we can improve the retrieval performance of a precedent retrieval system by indexing only the key concepts in the prior case documents. In addition, we investigate whether we could improve the retrieval performance by expanding the original queries and performing retrieval on a summarized document collection. The results suggest that expanding the current case can improve the retrieval performance when the retrieval is performed on a summarised document collection of prior cases. For statute retrieval, we investigate whether the retrieval performance could be improved by extracting only the key concepts from the queries or by expanding the queries without summarising the statute documents. The results of this study suggest that summarising the current case can improve the retrieval performance of a statute retrieval system.

In this study we investigate whether reducing very long Japanese Statute Law queries can improve the retrieval effectiveness of the Japanese civil law articles. In particular, we deploy two different query reduction techniques. In the first approach, we generate query terms by extracting key concepts from the original query using TagCrowd to create tag clouds. In order to test the generality of our approach, we deploy the Simplified Clarity Score query performance predictor to select the top 20 terms that are predicted to perform better during the retrieval. The results of this study suggest that reducing the original query can help improve the retrieval effectiveness of the Japanese civil law articles. In addition, expanding these reduced queries with more informative terms from the Japanese civil law articles result in the query drift problem and it degrades the retrieval performance.

In this article, it is hypothesized that personalizing the book search application by incorporating user profiles such as background of personal tastes, interests and previously seen books.can issue or produce a more effective query result set as well as an effective book recommendation.To meet this end, experiments were carried out to explore which topic representation gives the best result.Four different query representations, which are title, request, group and a combination of title-request-group were used.It was observed that the title-request-group query representation was best.In addition, an investigation was conducted to determine whether a learning to rank framework that incorporates topical relevance by exploiting user profiles for document re-ranking according to individual preference will issue a more effective result set.Moreover, an investigation was conducted to determine whether the use of keywords from profiles for query expansion and reformulation improves the search results.The results of these investigations suggest that a more effective query result set as well as an effective book recommendation can be attained by incorporating user profiles such as background of personal tastes, interests and previously seen books into the social book search application.

In this article, an empirical investigation was conducted to determine whether merging search results generated by multiple query variants with the same information need can improve the retrieval performance in patient centered health information retrieval.In addition, this approach was compared with the selective collection enrichment approach, where only the results generated by a single query, which was predicted to perform better on the local collection is used.Three different results merging strategies predominantly used in distributed search environments with large overlapping databases were used in this study.The results of this investigation suggests that merging results using multiple query variants with the same information need can improve the retrieval performance.Also it was observed that the choice of an external collection used in generating these query variants can have an impact in the retrieval performance as it can sometimes lead to a degradation in the retrieval performance.When a comparison was made between results merging strategies and the selective collection enrichment approach, it was observed that the selective collection enrichment approach ranks fewer and highly relevant documents in the top 10 retrieved documents while the results merging strategies ranks more and slightly relevant documents in the top 10 retrieved documents.

When developing an automated FAQ retrieval system, the information supplier constructs question candidates in advance using their own knowledge. Then they answer these question candidates to create question-answer pairs to use in the FAQ retrieval system. However, these question-answer pairs will not always satisfy the users’ information needs. When there is no relevant question–answer pair to a users’ query, such a user may submit various query reformulations browsing over the long results list and may abandon the search before their information need has been satisfied. Such users many never return to use the system again because of the inability of the system to return relevant question-answer pairs to their query. In order to alleviate this, modern automated FAQ retrieval systems use a Missing Content Query (MCQ) detection subsystem to detect those queries that do not have the relevant question–answer pair. In this article we conduct a review of the different approaches proposed in the literature for detecting these MCQs. In particular, we provide a comprehensive review of the different systems that deployed the binary classification approach, the thresholding approach and the hybrid approach in the detection of MCQs. Moreover, we describe the strength and weaknesses of each approach.

Due to the increasing amount of electronic information, a majority of laypeople (ordinary people with no professional medical knowledge) now rely on the Web to seek for health information for self-diagnosis. However, research has shown that current search engines are failing to deliver effective search results due to the inability of laypeople to formulate good queries because of lack of domain knowledge and unfamiliarity with the medical vocabulary and concepts. This article attempts to address this by proposing a Selective Collection Enrichment approach that uses three different external resources to enrich a user query, thus generating three different expanded queries. In addition, we deploy pre-retrieval query performance predictors to select an expanded query that is most likely to perform better when retrieving on a local collection being searched. Furthermore, we evaluate the effects of combining several pre-retrieval query performance predictors' scores using data fusion techniques for Selective Collection Enrichment. Our empirical evaluation shows marked improvement in the retrieval performance in terms of nDCG@10 when the Selective Collection Enrichment approach is deployed.

Video content retrieval just like information retrieval requires some pre-processing such as indexing, key-frame selection and most importantly accurate video shot boundary detection. Accurate detection of video shots give way for video information to be stored in a manner that will allow easy access. Several algorithms have been developed in this field of study and tested even at the TRECVID 2002, 2005 and 2007 tasks evaluation conferences. Challenges on accurate detection of these different types of video transitions have always been from large object and camera motions as well as fast zooming, flashlights, and change in luminance. These attributes differ from one video sequence to another and features of one video sequence cannot always match with features of another video. Therefore, in our work we use a video specific machine learning approach that leverages information from several shot boundary detection algorithms in order to improve the detection of the shot boundaries on a video sequence. Our results suggest that a classifier built from a combination of block-based motion estimation, RGB histogram based block-based cross-correlation coefficient and RGB histogram based sum of squared difference provided better results with an average F1 score of 0.752 on shot boundary detection for the seventeen videos tested. This proves that a combination of luminance and motion based algorithms improves the detection of video shot boundaries. We also found that the detection of shot cuts and gradual transitions can be improved by using features generated by several shot boundary detection algorithms.

This paper presents our working notes for FIRE 2017, Information Retrieval from Legal documents -Task 2 (Precedence retrieval). Common Law Systems around the world recognize the importance of precedence in Law. In making decisions, Judges are obliged to consult prior cases that had already been decided to ensure that there is no divergence in treatment of similar situations in different cases. Our approach was to investigate the effectiveness of using legal catchphrases in precedence retrieval. To improve retrieval performance, we incorporated term dependency in our retrieval. In addition, we investigate the effects of deploying query expansion on the retrieval performance. Our results show an improvement in the retrieval performance when we incorporate term dependence in scoring and ranking prior cases. However, we see a degradation in the retrieval performance when we deploy query expansion.

In this paper, we describe the methods deployed in the different runs submitted for our participation to the CLEF eHealth 2017 Task 3: Patient-Centered Information Retrieval, IRTask 1: ad-hoc search. Specifically, we deploy DPH term weighting model with explicit relevance feedback, where the expansion terms are selected from documents which were previously identified as relevant by assessors for each query. As improvement we deployed proximity search using both Full Dependence (FD) and Sequential Dependence (SD) variants of the Markov Random Fields and the Divergence From Randomness (DFR) based dependence models to re-rank documents, which have query terms in close proximity. In another approach, we deploy pseudo relevance feedback, where the expansion terms are selected from the top 3 ranked documents after a first pass retrieval. In addition, we deploy proximity search using the SD variant of the DFR based dependence model.