Abstract: Question answering (QA), one of the important applications of Natural Language Processing (NLP) aims to take the user questions and returned to the user with the answers. An open domain QA system deals with a set of questions that can be of any domain. The other type of QA is close-domain where it deals with the questions under a specific domain e.g., agriculture, medicine, education, tourism, etc. Our cooking question answering system is an example of a closed domain QA system. Here, users can ask the cooking related questions and the system returns the actual answer to the user. In this paper, we present different modules of a cooking QA system. In addition to dataset preparation, the development of a cooking ontology, the classification of questions as well as the extraction of candidate answers are also treated as other important aspects, which are discussed in this paper in details. In the cooking QA system, automatic evaluation metrics such as precision, recall, F-score, and C@1 were used for the evaluation of precise answers. In addition, human evaluation is used based on a rating scale. Moreover, the recommendation of recipes has also been attempted and the evaluation metrics show satisfactory performances of the systems.

Abstract: In an automated Question Answering (QA) system, Question Classification (QC) is an essential module. The aim of QC is to identify the type of questions and classify them based on the expected answer type. Although the machine-learning approach overcomes the limitation of rules as is the case with the conventional rule-based approach but is restricted to the predefined class of questions. The existing approaches are too specific for the users. To address this challenge, we have developed a cooking QA system in which a recipe question is contextually classified into a particular category using deep learning techniques. The question class is then used to extract the requisite details from the recipe obtained via the rule-based approach to provide a precise answer. The main contribution of this paper is the description of the QC module of the cooking QA system. The obtained intermediate classification accuracy over the unseen data is 90% and the human evaluation accuracy of the final system output is 39.33%.

Abstract: An evaluation metric is an absolute necessity for measuring the performance of any system and complexity of any data. In this paper, we have discussed how to determine the level of complexity of code-mixed social media texts that are growing rapidly due to multilingual interference. In general, texts written in multiple languages are often hard to comprehend and analyze. At the same time, in order to meet the demands of analysis, it is also necessary to determine the complexity of a particular document or a text segment. Thus, in the present paper, we have discussed the existing metrics for determining the code-mixing complexity of a corpus, their advantages and shortcomings as well as proposed several improvements on the existing metrics. The new index better reflects the variety and complexity of a multilingual document. Also, the index can be applied to a sentence and seamlessly extended to a paragraph or an entire document. We have employed two existing code-mixed corpora to suit the requirements of our study.

Abstract: Effective retrieval of mathematical contents from vast corpus of scientific documents demands enhancement in the conventional indexing and searching mechanisms. Indexing mechanism and the choice of semantic similarity measures guide the results of Math Information Retrieval system (MathIRs) to perfection. Tokenization and formula unification are among the distinguishing features of indexing mechanism, used in MathIRs, which facilitate sub-formula and similarity search. Besides, the scientific documents and the user queries in MathIRs will contain math as well as text contents and to match these contents we require three important modules: Text-Text Similarity (TS), Math-Math Similarity (MS) and Text-Math Similarity (TMS). In this paper we have proposed MathIRs comprising these important modules and a substitution tree based mechanism for indexing mathematical expressions. We have also presented experimental results for similarity search and argued that proposal of MathIRs will ease the task of scientific document retrieval.

Abstract Music information retrieval is currently an active domain of research. An interesting aspect of music information retrieval involves mood classification. While the Western music captured much attention, research on Indian music was limited and mostly based on audio data. In this work, the authors propose a mood taxonomy and describe the framework for developing a multimodal dataset (audio and lyrics) for Hindi songs. We observed differences in mood for several instances of Hindi songs while annotating the audio of such songs in contrast to their corresponding lyrics. Finally, the mood classification frameworks were developed for Hindi songs and they consist of three different systems based on the features of audio, lyrics and both. The mood classification systems based on audio and lyrics achieved F-measures of 58.2% and 55.1%, respectively whereas the multimodal system (combination of both audio and lyrics) achieved the maximum F-measure of 68.6%.

Movie scripts are a rich textual resource that can be tapped for movie content analysis. This article describes a mechanism for fragmenting a sequence of movie script dialogue into scene-wise groups. In other words, it attempts to locate scene transitions using information acquired from a sequence of dialogue units. We collect movie scripts from a web archive. Thereafter, we preprocess them to develop a resource of dialogues. We feed the dialogue sequence from a script to a Genetic Algorithm (GA) framework. The system fragments the sequence into adjacent groups of dialogue units or output 'scenes'. We use SentiWordnet scores and Wordnet distance for dialogue units to optimize this grouping so that adjacent scenes are semantically most dissimilar. Then we compare the resulting fragmented dialogue sequence with the original scene-wise alignment of dialogue in the script.

Zn-ferrite nanoparticles were synthesized at room temperature by mechanical alloying the stoichiometric (1:1 mol%) mixture of ZnO and α-Fe2O3 powder under open air. Formation of both normal and inverse spinel ferrite phases was noticed after 30 minutes and 2.5 hours ball milling respectively and the content of inverse spinel phase increased with increasing milling time. The phase transformation kinetics towards formation of ferrite phases and microstructure characterization of ball milled ZnFe2O4 phases was primarily investigated by X-ray powder diffraction pattern analysis. The relative phase abundances of different phases, crystallite size, r.m.s. strain, lattice parameter change etc. were estimated from the Rietveld powder structure refinement analysis of XRD data. Positron annihilation lifetime spectra of all ball milled samples were deconvoluted with three lifetime parameters and their variation with milling time duration was explained with microstructural changes and formation of different phases with increase of milling time duration.

The present paper describes the identification and tracking of bloggers' emotions with respect to time from the structured Bengali blog documents. The assignment of Ekman's six basic emotions to the bloggers' comments is carried out at sentence and paragraph level granularities. The Referential Informative Chain (RIC) developed for each blogger consists of the nodes representing the emotional states of that blogger. Each node of a RIC contains the identification information of its associated blogger, timestamp, section and emotional sentences. The nodes are arranged in each RIC based on the ascending order of the associated timestamps. An affect scoring technique has been employed to capture the emotions from each of the nodes of a blogger's RIC. The incorporation of self emotions and influential emotions as extracted from other bloggers plays a significant role in detecting the emotions of a blogger's present state. The Extrinsic evaluation produces precision (P), recall (R) and F-Measure of 61.05%, 69.81% and 65.13% respectively for evaluating the total of 193 emotional states of 20 bloggers. The Intrinsic evaluation has been conducted using a manual rater with the help of a statistical agreement coefficient, Krippendorff s alpha a. Two types of alpha, namely nominal alpha and interval alpha produce the average scores of 0.67 and 0.72, respectively.

The present task involves the identification of emotions from Bengali blog documents using two separate approaches. The first one is a machine learning approach that accumulates document level information from sentences obtained from word level granular detail whereas the second one is a resource based approach that considers the Bengali WordNet Affect, the word level Bengali affective lexical resource. In the first approach, the Support Vector Machine (SVM) classifier is employed to perform the word level classification. Sense weight based average scoring technique determines the sentential emotion scores based on the word level emotion tagged constituents. The cumulative summation of sentential emotion scores is assigned to each document considering the combinations of various heuristic features. The second one implements a majority based approach to classify a given document considering the Bengali WordNet Affect lists. Instead of assigning a single emotion tag to a document, in both approaches, the best two emotion tags are assigned to each document according to the ordered emotion scores obtained. By applying the best feature combination acquired from the development set, the evaluation of 110 test documents yields the average F-Scores of 59.50% and 51.07% for the two approaches respectively with respect to all emotion classes.

El objetivo de este trabajo es identificar las emociones en documentos escritos en bengalí extraídos de un blog usando dos enfoques distintos. El primer enfoque es aprendizaje automático en el cual se acumula la información de los documentos a partir de las oraciones obtenidas a través de análisis de palabras, es decir, en el nivel más granular, mientras que el segundo enfoque está basado en recursos de los cuales usamos el Bengalí WordNet Affect -un recurso léxico que incluye palabras del bengalí etiquetadas con emociones. En el primer enfoque, la máquina de soporte vectorial (Support Vector Machine, SVM) se usa para la clasificación a nivel de palabras. El valor afectivo de las oraciones se calcula según la técnica basada en promediar los puntajes de pesos asignados a los significados de palabras etiquetadas con emociones en estas oraciones. La suma acumulada de los puntajes afectivos de las oraciones se asigna a cada documento tomando en cuenta diversas características heurísticas. El segundo enfoque implementa el método basado en mayoría para clasificar un documento dado considerando las listas del Bengalí WordNet Affect. En ambos enfoques, en vez de asignar una única etiqueta afectiva a un documento dado, las dos mejores etiquetas afectivas se asignan a cada documento según los puntajes afectivos obtenidos ordenados. Usando la combinación de las mejores características obtenida del conjunto de desarrollo, al evaluar 110 documentos de prueba resulta un valor promedio de la métrica F-score en los dos enfoques 59,50% y 51,07% respectivamente para toda clase de emociones.