Mirna Adriani

Discovering high quality answers in community question answering archives using a hierarchy of classifiers

In community-based question answering (CQA) services where answers are generated by human, users may expect better answers than an automatic question answering system. However, in some cases, the user generated answers provided by CQA archives are not always of high quality. Most existing works on answer quality prediction use the same model for all answers, despite the fact that each answer is intrinsically different. However, modeling each individual QA pair differently is not feasible in practice. To balance between efficiency and accuracy, we propose a hybrid hierarchy-of-classifiers framework to model the QA pairs. First, we analyze the question type to guide the selection of the right answer quality model. Second, we use the information from question analysis to predict the expected answer features and train the type-based quality classifiers to hierarchically aggregate an overall answer quality score. We also propose a number of novel features that are effective in distinguishing the quality of answers. We tested the framework on a dataset of about 50 thousand QA pairs from Yahoo! Answer. The results show that our proposed framework is effective in identifying high quality answers. Moreover, further analysis reveals the ability of our framework to classify low quality answers more accurately than a single classifier approach.

Developing Non-goal Dialog System Based on Examples of Drama Television

This paper presents a design and experiments of developing a non-goal dialog system by utilizing human-to-human conversation examples from drama television. The aim is to build a conversational agent that can interact with users in as natural a fashion as possible, while reducing the time requirement for database design and collection. A number of the challenging design issues we faced are described, including (1) filtering and constructing a dialog example database from the drama conversations and (2) retrieving a proper system response by finding the best dialog example based on the current user query. Subjective evaluation from a small user study is also discussed.

Finding answers to indonesian questions from english documents

We present a report on our participation in the Indonesian-English question-answering task of the 2005 Cross-Language Evaluation Forum (CLEF). In this work we translated an Indonesian query set into English using a commercial machine translation tool called Transtool. We used linguistic tools to find the answer to a question. The answer is extracted from a relevant passage and is identified as having the relevant tagging as the query.

A Comparative Study on Twitter Sentiment Analysis: Which Features are Good?

In this paper, investigations of Sentiment Analysis over a well-known Social Media Twitter were done. As literatures show that some works related to Twitter Sentiment Analysis have been done and delivered interesting idea of features, but there is no a comparative study that shows the best features in performing Sentiment Analysis. In total we used 9 feature sets (41 attributes) that comprise punctuation, lexical, part of speech, emoticon, SentiWord lexicon, AFINN-lexicon, Opinion lexicon, Senti-Strength method, and Emotion lexicon. Feature analysis was done by conducting supervised classification for each feature sets and continued with feature selection in subjectivity and polarity domain. By using four different datasets, the results reveal that AFINN lexicon and Senti-Strength method are the best current approaches to perform Twitter Sentiment Analysis.

Query and document translation for english-Indonesian cross language IR

We present a report on our participation in the Indonesian-English ad hoc bilingual task of the 2006 Cross-Language Evaluation Forum (CLEF). This year we compare the use of several language resources to translate Indonesian queries into English. We used several readable machine dictionaries to perform the translation. We also used two machine translation techniques to translate the Indonesian queries. In addition to translating an Indonesian query set into English, we also translated English documents into Indonesian using the machine readable dictionaries and a commercial machine translation tool. The results show performing the task by translating the queries is better than translating the documents. Combining several dictionaries produced better result than only using one dictionary. However, the query expansion that we applied to the translated queries using the dictionaries reduced the retrieval effectiveness of the queries.

Buzzer Detection and Sentiment Analysis for Predicting Presidential Election Results in a Twitter Nation

In this paper, we present our approach for predicting the results of Indonesian Presidential Election using Twitter as our main resource. We explore the possibility of easy-togather Twitter data to be utilized as a survey supporting tool to understand public opinion. First, we collected Twitter data during the campaign period. Second, we performed automatic buzzer detection on our Twitter data to remove those tweets generated by computer bots, paid users, and fanatic users that usually become noise in our data. Third, we performed a fine-grained political sentiment analysis to partition each tweet into several sub-tweets and subsequently assigned each sub-tweet with one of the candidates and its sentiment polarity. Finally, to predict the election results, we leveraged the number of positive sub-tweets for each candidate. Our experiment shows that the mean absolute error (MAE) of our Twitter-based prediction is 0.61%, which is surprisingly better than the prediction results published by several independent survey institutions (offline polls). Our study suggests that Twitter can serve as an important resource for any political activity, specifically for predicting the final outcomes of the election itself.

Identifying location in indonesian documents for geographic information retrieval

Our research focuses on Geographic Information Retrieval for Indonesian documents. We constructed a Geographical Gazeeter for geographic locations based on information that we collected from the Geographic resources. We used the table to identify the locations of events that are mentioned in the documents. We added a location index associated with each document. We also identified the possible locations based on words that indicate the relative position to locations mentioned in the queries. The result of our experiments showed that the location index and the processing of words indicating relative positions improved the retrieval performance of natural language queries. Applying the query expansion techniques by adding keywords and locations from the documents to the queries resulted in significant performance improvement.

The Use of POS Sequence for Analyzing Sentence Pattern in Twitter Sentiment Analysis

As one of the largest Social Media in providing public data every day, Twitter has attracted the attention of researcher to investigate, in order to mine public opinion, which is known as Sentiment Analysis. Consequently, many techniques and studies related to Sentiment Analysis over Twitter have been proposed in recent years. However, there is no study that discuss about sentence pattern of positive/negative sentence and neither subjective/objective sentence. In this paper we propose POS sequence as feature to investigate pattern or word combination of tweets in two domains of Sentiment Analysis: subjectivity and polarity. Specifically we utilize Information Gain to extract POS sequence in three forms: sequence of 2-tags, 3-tags, and 5-tags. The results reveal that there are some tendencies of sentence pattern which distinguish between positive, negative, subjective and objective tweets. Our approach also shows that feature of POS sequence can improve Sentiment Analysis accuracy.

Combination of two-dimensional cochleogram and spectrogram features for deep learning-based ASR

This paper explores the use of auditory features based on cochleograms; two dimensional speech features derived from gammatone filters within the convolutional neural network (CNN) framework. Furthermore, we also propose various possibilities to combine cochleogram features with log-mel filter banks or spectrogram features. In particular, we combine within low and high levels of CNN framework which we refer to as low-level and high-level feature combination. As comparison, we also construct the similar configuration with deep neural network (DNN). Performance was evaluated in the framework of hybrid neural network - hidden Markov model (NN-HMM) system on TIMIT phoneme sequence recognition task. The results reveal that cochleogram-spectrogram feature combination provides significant advantages. The best accuracy was obtained by high-level combination of two dimensional cochleogram-spectrogram features using CNN, achieved up to 8.2% relative phoneme error rate (PER) reduction from CNN single features or 19.7% relative PER reduction from DNN single features.

Stochastic Gradient Variational Bayes for deep learning-based ASR

Many successful methods for training deep neural networks (DNN) rely on an unsupervised pretraining algorithm. It is particularly effective when the number of labeled training samples is not large enough, because pretraining method helps to initialize the parameter values in the appropriate range near a local good minimum, for further discriminative finetuning. However, while the improvement is impressive, training DNN is difficult because the objective function of DNN is highly non-convex function of the parameters. To avoid placing the parameter that generalizes poorly, a robust generative modelling is necessary. This paper explore an alternative of generative modelling for pretraining DNN-based acoustic modelling using Stochastic Gradient Variational Bayes (SGVB) within autoencoder framework called Variational Bayes Autoencoder (VBAE). It performs an efficient approximate inference and learning with directed probabilistic graphical models. During fine-tuning, probabilistic encoder parameters with latent variable components are then used in discriminative training for acoustic model. Here, we investigate the performances of DNN-based acoustic model using the proposed pretrained VBAE in comparison with widely used pretraining algorithms like Restricted Boltzmann Machine (RBM) and Stacked Denoising Autoencoder (SDAE). The results reveal that VBAE pretraining with Gaussian latent variables gave the best performance.