Fidel Cacheda

Comparison of collaborative filtering algorithms: Limitations of current techniques and proposals for scalable, high-performance recommender systems

The technique of collaborative filtering is especially successful in generating personalized recommendations. More than a decade of research has resulted in numerous algorithms, although no comparison of the different strategies has been made. In fact, a universally accepted way of evaluating a collaborative filtering algorithm does not exist yet. In this work, we compare different techniques found in the literature, and we study the characteristics of each one, highlighting their principal strengths and weaknesses. Several experiments have been performed, using the most popular metrics and algorithms. Moreover, two new metrics designed to measure the precision on good items have been proposed. The results have revealed the weaknesses of many algorithms in extracting information from user profiles especially under sparsity conditions. We have also confirmed the good results of SVD-based techniques already reported by other authors. As an alternative, we present a new approach based on the interpretation of the tendencies or differences between users and items. Despite its extraordinary simplicity, in our experiments, it obtained noticeably better results than more complex algorithms. In fact, in the cases analyzed, its results are at least equivalent to those of the best approaches studied. Under sparsity conditions, there is more than a 20% improvement in accuracy over the traditional user-based algorithms, while maintaining over 90% coverage. Moreover, it is much more efficient computationally than any other algorithm, making it especially adequate for large amounts of data.

Extracting lists of data records from semi-structured web pages

Many web sources provide access to an underlying database containing structured data. These data can be usually accessed in HTML form only, which makes it difficult for software programs to obtain them in structured form. Nevertheless, web sources usually encode data records using a consistent template or layout, and the implicit regularities in the template can be used to automatically infer the structure and extract the data. In this paper, we propose a set of novel techniques to address this problem. While several previous works have addressed the same problem, most of them require multiple input pages while our method requires only one. In addition, previous methods make some assumptions about how data records are encoded into web pages, which do not always hold in real websites. Finally, we have also tested our techniques with a high number of real web sources and we have found them to be very effective.

Twitter: A Good Place to Detect Health Conditions

With the proliferation of social networks and blogs, the Internet is increasingly being used to disseminate personal health information rather than just as a source of information. In this paper we exploit the wealth of user-generated data, available through the micro-blogging service Twitter, to estimate and track the incidence of health conditions in society. The method is based on two stages: we start by extracting possibly relevant tweets using a set of specially crafted regular expressions, and then classify these initial messages using machine learning methods. Furthermore, we selected relevant features to improve the results and the execution times. To test the method, we considered four health states or conditions, namely flu, depression, pregnancy and eating disorders, and two locations, Portugal and Spain. We present the results obtained and demonstrate that the detection results and the performance of the method are improved after feature selection. The results are promising, with areas under the receiver operating characteristic curve between 0.7 and 0.9, and f-measure values around 0.8 and 0.9. This fact indicates that such approach provides a feasible solution for measuring and tracking the evolution of health states within the society.

Experiences retrieving information in the world wide web

In this paper using the information obtained from the daily working of a Web directory, we attempt to expand the knowledge about the behavior of the users in order to improve and adapt the Internet search engines to their users. We have analysed more than 320,000 requests of the transaction log of a Spanish Web directory, focusing our attention, firstly, in the searches in order to confirm the main differences between Internet and traditional Information Retrieval systems. Furthermore, we have developed an exhaustive statistical analysis of searches, categories visited and documents viewed to achieve a mathematical pattern of behaviour for each one, and what it is more important, to establish a relationship between the variations in the behaviour of each one.

Performance analysis of distributed information retrieval architectures using an improved network simulation model

The increasing number of documents that have to be indexed in different environments, particularly on the Web, and the lack of scalability of a single centralised index lead to the use of distributed information retrieval systems to effectively search for and locate the required information. In this study, we present several improvements over the two main bottlenecks in a distributed information retrieval system (the network and the brokers). We extend a simulation network model in order to represent a switched network. The new simulation model is validated by comparing the estimated response times with those obtained using a real system. We show that the use of a switched network reduces the saturation of the interconnection network, especially in a replicated system, and some improvements may be achieved using multicast messages and faster connections with the brokers. We also demonstrate that reducing the partial results sets will improve the response time of a distributed system by 53%, with a negligible probability of changing the systems precision and recall values. Finally, we present a simple hierarchical distributed broker model that will reduce the response times for a distributed system by 55%.

Performance Analysis of Distributed Architectures to Index One Terabyte of Text

We simulate different architectures of a distributed Information Retrieval system on a very large Web collection, in order to work out the optimal setting for a particular set of resources. We analyse the effectiveness of a distributed, replicated and clustered architecture using a variable number of workstations. A collection of approximately 94 million documents and 1 terabyte of text is used to test the performance of the different architectures. We show that in a purely distributed architecture, the brokers become the bottleneck due to the high number of local answer sets to be sorted. In a replicated system, the network is the bottleneck due to the high number of query servers and the continuous data interchange with the brokers. Finally, we demonstrate that a clustered system will outperform a replicated system if a large number of query servers is used, mainly due to the reduction of the network load.

Finding and Extracting Data Records from Web Pages

Many HTML pages are generated by software programs by querying some underlying databases and then filling in a template with the data. In these situations the metainformation about the data structure is lost, so automated software programs cannot process these data in such powerful manners as information from databases. We propose a set of novel techniques for detecting structured records in a web page and extracting the data values that constitute them. Our method needs only an input page. It starts by identifying the data region of interest in the page. Then it is partitioned into records by using a clustering method that groups similar subtrees in the DOM tree of the page. Finally, the attributes of the data records are extracted by using a method based on multiple string alignment. We have tested our techniques with a high number of real web sources, obtaining high precision and recall values.

A self-adapting latency/power tradeoff model for replicated search engines

For many search settings, distributed/replicated search engines deploy a large number of machines to ensure efficient retrieval. This paper investigates how the power consumption of a replicated search engine can be automatically reduced when the system has low contention, without compromising its efficiency. We propose a novel self-adapting model to analyse the trade-off between latency and power consumption for distributed search engines. When query volumes are high and there is contention for the resources, the model automatically increases the necessary number of active machines in the system to maintain acceptable query response times. On the other hand, when the load of the system is low and the queries can be served easily, the model is able to reduce the number of active machines, leading to power savings. The model bases its decisions on examining the current and historical query loads of the search engine. Our proposal is formulated as a general dynamic decision problem, which can be quickly solved by dynamic programming in response to changing query loads. Thorough experiments are conducted to validate the usefulness of the proposed adaptive model using historical Web search traffic submitted to a commercial search engine. Our results show that our proposed self-adapting model can achieve an energy saving of 33% while only degrading mean query completion time by 10 ms compared to a baseline that provisions replicas based on a previous days traffic.

Hybrid query scheduling for a replicated search engine

Search engines use replication and distribution of large indices across many query servers to achieve efficient retrieval. Under high query load, queries can be scheduled to replicas that are expected to be idle soonest, facilitated by the use of predicted query response times. However, the overhead of making response time predictions can hinder the usefulness of query scheduling under low query load. In this paper, we propose a hybrid scheduling approach that combines the scheduling methods appropriate for both low and high load conditions, and can adapt in response to changing conditions. We deploy a simulation framework, which is prepared with actual and predicted response times for real Web search queries for one full day. Our experiments using different numbers of shards and replicas of the 50 million document ClueWeb09 corpus show that hybrid scheduling can reduce the average waiting times of one day of queries by 68% under high load conditions and by 7% under low load conditions w.r.t. traditional scheduling methods.

Using clustering and edit distance techniques for automatic web data extraction

Many web sources provide access to an underlying database containing structured data. These data can be usually accessed in HTML form only, which makes it difficult for software programs to obtain them in structured form. Nevertheless, web sources usually encode data records using a consistent template or layout, and the implicit regularities in the template can be used to automatically infer the structure and extract the data. In this paper, we propose a set of novel techniques to address this problem. While several previous works have addressed the same problem, most of them require multiple input pages while our method requires only one. In addition, previous methods make some assumptions about how data records are encoded into web pages, which do not always hold in real websites. Finally, we have tested our techniques with a high number of real web sources and we have found them to be very effective.