Ahmed Hefny

A nonparametric mixture model for topic modeling over time

A single, stationary topic model such as latent Dirichlet allocation is inappropriate for modeling corpora that span long time periods, as the popularity of topics is likely to change over time. A number of models that incorporate time have been proposed, but in general they either exhibit limited forms of temporal variation, or require computationally expensive inference methods. In this paper we propose nonparametric Topics over Time (npTOT), a model for time-varying topics that allows an unbounded number of topics and flexible distribution over the temporal variations in those topics’ popularity. We develop a collapsed Gibbs sampler for the proposed model and compare against existing models on synthetic and real document sets.

Incremental combinatory categorial grammar and its derivations

Incremental parsing is appealing for applications such as speech recognition and machine translation due to its inherent efficiency as well as being a natural match for the language models commonly used in such systems. In this paper we introduce an Incremental Combinatory Categorical Grammar (ICCG) that extends the standard CCG grammar to enable fully incremental left-to-right parsing. Furthermore, we introduce a novel dynamic programming algorithm to convert CCGbank normal form derivations to incremental left-to-right derivations and show that our incremental CCG derivations can recover the unlabeled predicate-argument dependency structures with more than 96% F-measure. The introduced CCG incremental derivations can be used to train an incremental CCG parser.

Exploring friend's influence in cultures in Twitter

What does a user do when he logs in to the Twitter website? Does he merely browse through the tweets of all his friends as a source of information for his own tweets, or does he simply tweet a message of his own personal interest? Does he skim through the tweets of all his friends or only of a selected few? A number of factors might influence a user in these decisions. Does this social influence vary across cultures? In our work, we propose a simple yet effective model to predict the behavior of a user - in terms of which hashtag or named entity he might include in his future tweets. We have approached the problem as a classification task with the various influences contributing as features. Further, we analyze the contribution of the weights of the different features. Using our model we analyze data from different cultures and discover interesting differences in social influence.

Is a query worth translating: ask the users!

Users in many regions of the world are multilingual and they issue similar queries in different languages. Given a source language query, we propose query picking which involves finding equivalent target language queries in a large query log. Query picking treats translation as a search problem, and can serve as a translation method in the context of cross-language and multilingual search. Further, given that users usually issue queries when they think they can find relevant content, the success of query picking can serve as a strong indicator to the projected success of cross-language and multilingual search. In this paper we describe a system that performs query picking and we show that picked queries yield results that are statistically indistinguishable from a monolingual baseline. Further, using query picking to predict the effectiveness of cross-language results can have statistically significant effect on the success of multilingual search with improvements over a monolingual baseline. Multilingual merging methods that do not account for the success of query picking can often hurt retrieval effectiveness.

Query-based Workload Forecasting for Self-Driving Database Management Systems

The first step towards an autonomous database management system (DBMS) is the ability to model the target applications workload. This is necessary to allow the system to anticipate future workload needs and select the proper optimizations in a timely manner. Previous forecasting techniques model the resource utilization of the queries. Such metrics, however, change whenever the physical design of the database and the hardware resources change, thereby rendering previous forecasting models useless. We present a robust forecasting framework called QueryBot 5000 that allows a DBMS to predict the expected arrival rate of queries in the future based on historical data. To better support highly dynamic environments, our approach uses the logical composition of queries in the workload rather than the amount of physical resources used for query execution. It provides multiple horizons (short- vs. long-term) with different aggregation intervals. We also present a clustering-based technique for reducing the total number of forecasting models to maintain. To evaluate our approach, we compare our forecasting models against other state-of-the-art models on three real-world database traces. We implemented our models in an external controller for PostgreSQL and MySQL and demonstrate their effectiveness in selecting indexes.

A new monte carlo-based error rate estimator

Estimating the classification error rate of a classifier is a key issue in machine learning. Such estimation is needed to compare classifiers or to tune the parameters of a parameterized classifier. Several methods have been proposed to estimate error rate, most of which rely on partitioning the data set or drawing bootstrap samples from it. Error estimators can suffer from bias (deviation from actual error rate) and/or variance (sensitivity to the data set). In this work, we propose an error rate estimator that estimates a generative and a posterior probability models to represent the underlying process that generates the data and exploits these models in a Monte Carlo style to provide two biased estimators whose best combination is determined by an iterative solution. We test our estimator against state of the art estimators and show that it provides a reliable estimate in terms of mean-square-error.

Fast and Improved SLEX Analysis of High-Dimensional Time Series

We address the problem of segmenting a multi-dimensional time series into stationary blocks by improving AutoSLEX [1], which has been successfully used for this purpose. AutoSLEX finds the best basis in a library of smoothed localized exponentials (SLEX) basis functions that are orthogonal and localized in both time and frequency. We introduce DynamicSLEX, a variant of AutoSLEX that relaxes the dyadic intervals constraint of AutoSLEX, allowing for more flexible segmentation while maintaining tractability. Then, we introduce RandSLEX, which uses random projections to scale-up SLEX-based segmentation to high dimensional inputs and to establish a notion of strength of splitting points in the segmentation. We demonstrate the utility of the proposed improvements on synthetic and real data.