Archivo

Posts Tagged ‘Big data’

Analysis: Modeling Air Pollution in the city of Santander (Spain)

We have published a new study entitled “Modeling Air Pollution in the City of Santander (Spain)“, carried out in the context of the project Ciudad2020. In this new document – in a similar way to what we did in our study on noise pollution-, we have focused on presenting the full analysis of real application in the modeling of air pollution in the city of Santander (Spain), which had already been summarily described in our whitepaper on pollution predictive modeling techniques in the sustainable city.

One of the objectives of Ciudad2020 as far as pollution in concerned is to install across the city a wide network of low-cost sensors (with respect to the current model, made of few very expensive and accurate measuring stations). However, at present, the mentioned low-cost sensor network has not been deployed in any city yet, and checking the validity of this model requires data about various pollutants related to an urban center.

cimaThe data used in this analysis are historical data provided by the Environmental Research Centre (CIMA).This entity is an autonomous body of the Government of Cantabria created by law in 1991 and headed by the Ministry of Environment. Its activity is centered on the realization of physico-chemical analyses on the state of the environment and the management of sustainability through Environmental Information, Participation, Education and Environmental Volunteering.

The data set consists of measures taken every 15 minutes between 1/1/2011 and 31/1/2013 by 4 automatic measuring stations of the Air Quality Control and Monitoring Network of Cantabria, which are located in the surroundings of Santander. The values associated to pollutants are the following: PM10 (particles in suspension of size less than 10 microns), SO2 (sulphur dioxide), NO and NO2 (nitrogen oxides), CO (carbon monoxide), O3 (ozone), BEN (benzene), TOL (toluene) and XIL (xylene). In addition, those stations that have a meteorological tower measure the following meteorological parameters: DD (wind direction), VV (wind speed), TMP (temperature), HR (relative humidity), PRB (atmospheric pressure), RS (solar radiation) and LL (precipitation level).

As described in the document, the first step in any modeling study consists in the analysis of data, performed variable by variable and from each measuring station. At least a study of the basic statistics by season (average and standard deviation, median, mode), the distribution of values (histogram) both at global and monthly level and the hourly distribution are requested. The moving average is also analyzed, a statistical feature applicable to the analysis of tendencies which smoothes the fluctuations typical of instant measurements and captures the trends in a given period.

estaciones-cantabria

The next step is to analyze how the variables depend on the others, in order to select the set of variables that most governs the behavior of the output variable. For that purpose correlation analysis has been employed, which is a statistical tool that allows measuring and describing the degree or intensity of association between two variables. In particular, Pearson’s correlation coefficient has been used, which measures the linear relationship between two random quantitative variables X and Y.

Analyses of dependencies have been carried out at the same moment of time, in moments of the past, with differentiated values (difference between the concentration level registered for a contaminant in a given moment of time and the level of 30 minutes before, aiming at detecting trends over time regardless of absolute values) and the moving average value of such contaminant considering different time intervals.

The next step is to evaluate a series of algorithms of modeling with monitored learning (prediction, classification) or not monitored (grouping) to draw conclusions about the behavior of pollution variables. The prediction analysis has been focused on Santander’s center, with 1-hour, 2-hour, 4-hour, 8-hour and 24-hour prediction horizons. Then, the models for each pollution variable in all those horizons have been trained and evaluated. Different machine learning algorithms have been trained in each case (variable-prediction horizon combination): M5P, IBk, Multilayer Perceptron, linear regression, Regression by Discretization, RepTree, Bagging with RepTree, etc. The assessment is performed by comparing the mean absolute error of all different prediction methods.

pollution

For example, when studying the 8-hour prediction, it can be noticed that the hour of the day becomes more important, since citizens behave cyclically and probably what happens at 7 a.m. (e.g. people go to work) relates to what happens at 3 p.m. (e.g. people come back from work).

The last step of the data mining process according to the CRISP-DM methodology would be the implementation in a system of environmental management for obtaining real-time predictions on the different values of pollutants. This implementation has to consider logically the results and conclusions obtained in the analysis and modeling processes at the time of setting up the deployment and prioritizing possible investments.

The most important thing to emphasize is that the analysis illustrates and details the steps to follow in a project of environmental pollution modeling using data mining, although, logically, the analysis and the concrete conclusions only apply, in general, to the city of Santander. You can access the complete study, more information and demos on our website: http://www.daedalus.es/ciudad2020/. If you have any questions or comments, please do not hesitate to contact us, we will be happy to assist you.

[Translation by Luca de Filippis]

Whitepaper: “Pollution Predictive Modeling in the Sustainable City”

Recently we have published the whitepaper “Pollution Predictive Modeling in the Sustainable City“, which describes in detail the approach and methodology that we have adopted within the framework of the Ciudad2020 project to perform predictive modeling of environmental pollution levels in the city of the future. Given that the starting point of the analysis is made up of the immense volume of data collected by the network of sensors deployed around the city, both physical sensors and the citizen sensor, this modeling is addressed as a data mining project (data analytics). Therefore, the methodology, techniques and algorithms typical of data mining have been used to process and exploit the information.

crispdmThe term KDD (Knowledge Discovery in Databases) was coined to refer to the (broad) concept of finding knowledge in data and to emphasize the high level application of certain data mining processes. In an attempt at normalizing this process of knowledge discovering, similarly to what it is done in software engineering for standardizing software development, two main methodologies were taken into account: SEMMA and CRISP-DM. Both fix the tasks to perform in each phase described by KDD, assigning specific tasks and defining the expected outcome for each phase. In (Azevedo, A. and Santos, M. F. KDD, SEMMA and CRISP-DM: a parallel overview. In Proceedings of the IADIS European Conference on Data Mining 2008, pp 182-185.), both implementations are compared and the conclusion is that, although you can draw a parallel between them, CRISP-DM is more complete. In fact, it takes into account also the application of outcomes to the business environment, and, for this reason, it has been adopted for modeling in Ciudad2020.

By collecting different documentary references, the whitepaper presents a detailed description of the CRISP-DM methodology, its objectives, essential phases and tasks. Then, it focuses on describing thoroughly the two application scenarios that have been considered in Ciudad2020 and the pollution modeling process carried out following this methodology: air pollution prediction in the city of Santander (Cantabria, Spain) and noise pollution prediction in the city of Madrid (Spain).

SERENA project (Spanish acronym for Neural Network Statistical Prediction System for Madrid’s Air Quality)

You can find the whitepaper, further information, more documentation and demos on our web page: http://www.daedalus.es/ciudad2020/. If you have any questions or comments, please do not hesitate to contact us, we will be happy to assist you.

[Translation by Luca de Filippis]

Entendiendo la TV Social con tecnologías de Análisis Semántico y Big Data

25 noviembre, 2013 Deja un comentario

Recientemente hemos participado en la conferencia Big Data Spain con una charla titulada “Real time semantic search engine for social TV streams”. En esta charla resumimos nuestra experiencia en el área de TV Social, donde estamos combinando tecnologías de análisis semántico y de procesamiento de flujos de datos (streams) en tiempo real para entender las redes sociales. 

La TV Social es un fenómeno en claro crecimiento ya que cada vez es más frecuente el uso de redes sociales mientras vemos la televisión. Por ejemplo, Twitter ya reportaba el año pasado que en Reino Unido más de un tercio de todos los comentarios durante el primetime eran sobre lo que pasaba en la pantalla del televisor. Hace semanas, Facebook reivindicaba su lugar en la TV social afirmando que el volumen de comentarios privados sobre TV en su plataforma era 5 veces mayor. Esta red social ha empezado también a ofrecer hashtags e incluso una nueva API, Keywords Insight, para que algunos socios tengan acceso a estadísticas agregadas de las conversaciones dentro de los muros de Facebook.

A medida que el número de usuarios que acaban comentando su programa favorito de TV en las redes sociales con amigos o extraños, las cadenas han empezado a hacer uso de estas herramientas para participar en la conversación. Durante el último año se ha incrementado el número de programas que hacen uso de hashtags oficiales, a veces incluso varios durante una sola emisión. Más allá de la búsqueda del trending topic, los que con mayor éxito han experimentado fomentan incluso la participación de presentadores o actores durante la emisión del programa. En España, “Salvados” o “Pesadilla en la Cocina” son ejemplo de lo primero y la serie “Isabel” para el caso de los actores.   

Aunque no hay nada nuevo en el hecho de socializar alrededor del contenido de la tele, la posibilidad de medir y de destilar estos comentarios sí que es característico del nuevo contexto. Además, esta conversación no se produce al día siguiente sino que su impacto es inmediato. Todo esto se suma para abrir un nuevo abanico de posibilidades para espectadores, radiodifusores y las grandes marcas. Los usuarios han encendido la mecha de la TV Social ya que les conecta con amigos y el resto de la audiencia.  Es a la vez una forma de sentirse parte del programa y al mismo tiempo de involucrarse más. Por tanto, las herramientas que faciliten organizar y comprender la conversación son de especial interés para el espectador social. Para el resto de espectadores, incluso para aquellos que solo participan leyendo, es una forma de generar tanto recomendaciones sociales. Otro uso interesante es que analizar la conversación social permite contextualizar contenido relevante y relacionado con el programa como información sobre los actores, personajes o cualquier otro concepto del que se hable en la pantalla grande.

Por otro lado, comprender la conversación en torno a un programa es de tremenda utilidad para los canales de TV y las marcas que los financian. Las productoras y cadenas pueden medir las preferencias de sus espectadores y los de la competencia, y además en tiempo real, según se está emitiendo. Está información cualitativa permite hacer una lectura mucho más informada de los datos cuantitativos de audiencia. Llegar a los consumidores adecuados y medir el impacto de las campañas de publicidad son solo dos ejemplos de lo que las marcas buscan en la TV social. Por último, no solo se trata de escuchar pasivamente, tanto los programas como los anuncios van a ser cada vez más interactivos y a buscar la participación de los espectadores mediante las redes sociales.  

En nuestra charla, describimos un sistema que combina análisis semántico y tecnologías de big data como una herramienta para sacar partido de las redes sociales. El sistema combina varios componentes de procesamiento de lenguaje natural de Textalytics API junto a una base de datos semi-estructurada, SenseiDB, para proporcionar visualizaciones interactivas de los comentarios de TV sobre la base de la búsqueda semántica, la búsqueda por facetas y un sistemas de analítica en tiempo real.

 

Mediante el uso de Textalytics API somos capaces de extraer información relevante para la TV social como el sentimiento acerca de una entidad (un programa, actor o deportista) Además, el reconocimiento de entidades y la extracción de información temática nos permite producir trending topics dentro de un programa con una alta correlación con lo que ocurre en pantalla. Estos temas unidos a las facetas online proporcionadas por SenseiDB funcionan como una manera efectiva de organizar la conversación al vuelo. Otras funcionalidades como el reconocimiento de idioma o la clasificación de texto cumplen un papel importante pero oculto y nos ayudan a filtrar el flujo de comentarios de contenido ruidoso.  

El segundo de los componentes esenciales del sistema es SenseiDB, una base de datos semi-estructurada de código abierto que permite ingerir streams y buscarlos en tiempo real, es decir, con baja latencia tanto en la indexación como en la búsqueda. SenseiDB incluye un gran número de tipos de facetas que permiten organizar los metadatos semánticos que extraemos con Textalytics. Con la ayuda de facetas de tipo histograma o rango se pueden realizar incluso ciertas tareas de analítica que incluyen tipos de datos complejos como el tiempo. Además, una de las características más interesantes es que incluye un lenguaje de consulta sencillo e intuitivo, BQL, que es de gran utilidad para el desarrollo iterativo de visualizaciones.

Si te ha parecido interesante, te animo a que le eches un vistazo a la presentación o incluso al vídeo del evento.

Semantic Analysis and Big Data to understand Social TV

25 noviembre, 2013 1 comentario

We recently participated in the Big Data Spain conference with a talk entitled “Real time semantic search engine for social TV streams”. This talk describes our ongoing experiments on Social TV and combines our most recent developments on using semantic analysis on social networks and dealing with real-time streams of data.

Social TV, which exploded with the use of social networks while watching TV programs is a growing and exciting phenomenon. Twitter reported that more than a third of their firehose in the primetime is discussing TV (at least in the UK) while Facebook claimed 5 times more comments behind his private wall. Recently Facebook also started to offer hashtags and the Keywords Insight API for selected partners as a mean to offer aggregated statistics on Social TV conversations inside the wall.

As more users have turned into social networks to comment with friends and other viewers, broadcasters have looked into ways to be part of the conversation. They use official hashtags, let actors and anchors to tweet live and even start to offer companion apps with social share functionalities.

While the concept of socializing around TV is not new, the possibility to measure and distill the information around these interactions opens up brand new possibilities for users, broadcasters and brands alike.  Interest of users already fueled Social TV as it fulfills their need to start conversations with friends, other viewers and the aired program. Chatter around TV programs may help to recommend other programs or to serve contextually relevant information about actors, characters or whatever appears in TV.  Moreover, better ways to access and organize public conversations will drive new users into a TV program and engage current ones.

On the other hand, understanding the global conversation about a program is definitely useful to acquire insights for broadcasters and brands. Broadcasters and TV producers may measure their viewers preferences and reactions or their competence and acquire complementary information beyond plain audience numbers. Brands are also interested in finding the most appropriate programs to reach their target users as well as understand the impact and acceptance of their ads. Finally, new TV and ad formats are already being created based on interaction and participation, which again bolster engagement.

In our talk, we describe a system that combines natural language processing components from our Textalytics API and a scalable semi-structured database/search engine, SenseiDB, to provide semantic and faceted search, real-time analytics and support visualizations for this kind of applications.

Using Textalytics API we are able to include interesting features for Social TV like analyzing the sentiment around an entity (a program, actor or sportsperson). Besides, entity recognition and topic extraction allow us to produce trending topics for a program that correlate well with whatever happens on-screen. They work as an effective form to organize the conversation in real-time when combined with the online facets provided by SenseiDB. Other functionalities like language recognition and text classification help us to clean the noisy streams of comments.

SenseiDB is the second pillar of our system. A semi-structured distributed database that helps us to ingest streams and made them available for search in real-time with low query and indexing times. It includes a large number of facet types that enable us to use navigation using a range of semantic information. With the help of histogram and range facets it could even be overused for simple analytics tasks. It is well rounded with a simple and elegant query language, BQL, which help us to boost the development of visualizations on top.

If you find it interesting, check out our presentation for more detail or even the video of the event.

Últimas tendencias en análisis de datos en Big Data Spain 2013

19 noviembre, 2013 Deja un comentario

logo Big Data SpainLa segunda edición de Big Data Spain, uno de los eventos pioneros en las tecnologías y aplicaciones del procesamiento masivo de datos se celebró el 7 y el 8 de Noviembre en Madrid. El evento que consiguió atraer a más de 400 asistentes, el doble que el año pasado, refleja el creciente interés en estas tecnologías también en España. Daedalus participó con una ponencia donde demostraba el uso de tecnologías de procesamiento de lenguaje natural, Big Data y redes sociales para el análisis en tiempo real de la TV social.

La tecnología de Big Data ha crecido y madurado cuando están a punto de cumplirse 10 años desde la publicación de MapReduce, el modelo de computación masiva y distribuida que marcó su inicio.

Rubén Casado, en una de las charlas más útiles para establecer un mapa del ingente número de proyectos de Big Data y NoSQL definió la historia de la tecnología en tres fases:

  • Procesamiento masivo en batch ( 2003 – ) con exponentes como Hadoop o Cassandra.
  • Procesamiento en tiempo real ( 2010 – ) representado con tecnologías como StormKafka o Samza
  • Procesamiento híbrido ( 2013 – ) que trata de unificar los dos anteriores en un modelo de programación única. Son ejemplos notables Summingbird  o Lambdoop.

Sin duda, la primera hornada de soluciones está lista para la empresa con distribuciones basadas en la pila tecnológica de Hadoop como Cloudera, MapR o HortonWorks. Del mismo modo crece el número de empresas que están integrando u ofrecen servicios de consultoría sobre Big Data en sectores diversos como banca, finanzas, telecomunicaciones o marketing.

Otras tres tendencias claras a nivel tecnológico son:

  • la popularización de sistemas que facilitan la analítica online de grandes volúmenes de datos (Spark, Impala, SploutSQL o SenseiDB)
  • la vuelta de SQL, o al menos de dialectos que reduzcan el tiempo de desarrollo
  • la importancia de la visualización como herramienta para comunicar los resultados de manera efectiva.

Pero, por supuesto, adoptar la filosofía Big Data en una empresa no es una cuestión puramente tecnológica. Requiere de una visión clara de los beneficios que genera basar tu negocio en datos y del valor y el conocimiento que se puede extraer integrando los datos internos y externos. Otro factor importante es contar con profesionales que sepan romper la barrera entre los aspectos más técnicos y los de negocio. En ese sentido cobra especial importancia la figura del científico de datos. Sean Owen de Cloudera la definió como “una persona que entiende la estadística mejor que un ingeniero software y es mejor en ingeniería software que cualquier estadístico”. Sin duda a estas habilidades hay que añadir el conocimiento del negocio y la capacidad para plantear las preguntas adecuadas.

Aunque no todas las opiniones coincidían, la mejor forma de empezar a “hacer Big Data” es poco a poco y abordando proyectos con objetivos de negocio bien definidos. Buenos candidatos para experimentar con la tecnología son aquellos procesos que ya suponen un cuello de botella. En otros casos, sin embargo, la necesidad viene por el lado de innovar, bien mediante la integración de datos externos o el diseño de productos basados en los datos. Buen ejemplo de este caso es la iniciativa de Big Data desde el Centro de Innovación BBVA que proporciona información agregada sobre transacciones de tarjetas de crédito.

TextalyticsPor último, y entroncando con lo que fue nuestra charla, uno de los tipos de fuentes externas donde hay un valor importante es en el uso de datos de las redes sociales. Por su heterogeneidad, se trata de uno de las fuentes de datos que plantea mas retos. Por esta razón, las herramientas de análisis de texto, como Textalytics API, deben formar parte de cualquier estrategia de Big Data ya que nos van a facilitar cruzar información cuantitativa y cualitativa con todo el valor que esto genera.

Si te interesa entrar en más profundidad, los videos de las charlas y los paneles de expertos se encuentran disponibles desde la web de Big Data Spain

Trends in data analysis from Big Data Spain 2013

19 noviembre, 2013 Deja un comentario

logo Big Data Spain

The second edition of Big Data Spain took place in Madrid on last November 7 and 8 and proved to be a landmark event on technologies and applications of big data processing. The event attracted more than 400 participants, doubling last year’s number, and reflected the growing interest on these technologies in Spain and across Europe. Daedalus participated with a talk that illustrated the use of natural language processing and Big Data technologies to analyze in real time the buzz around Social TV.

Big Data technology has matured when we are about to cellebrate its 10th birthday, marked by the publication of the MapReduce computing abstraction that later gave rise to the field.

Rubén Casado, in one of the most useful talks to understand the vast amnount of Big Data and NoSQL project outlined the recent history of the technology in three eras:

  • Batch processing ( 2003 – ) with examples like  Hadoop or Cassandra.
  • Real time processing ( 2010 – ) represented by recent projects like StormKafka o Samza.
  • Hybrid processing ( 2013 – ) which attempts to combine both worlds in an unified programming model like Summingbird  or Lambdoop.

Withouth any doubt, the first era of solutions is enterprise-ready with several Hadoop based distributions like Cloudera, MapR or HortonWorks. Likewise the number of companies that are integrating them or providing consultancy in this field is expanding and reaching every sector from finance and banking to telecomunications or marketing.

Some other technological trends clearly emerged from talk topics and panels:

  • growing number of alternatives to deal online with large volume data analysis tasks (Spark, Impala, SploutSQL o SenseiDB)
  • SQL comeback, or at least as dialects on top of actual systems that made easier to develop and maintain applications
  • the importance of visualization as a tool to communicate Big Data results effectively.

However, adopting Big Data as a philosophy inside your company is not just merely technology. It requires a clear vision of the benefits that grounding all your processes in data may carry, and the value and knowledge that you may obtain by integrating internal and also external data. Another important factor is to be able to find the right people to bridge the chasm between the technical and businness sides. In this sense, the role of the data scientist is very important and Sean Owen from Cloudera defined it as “a person who is better at statistics than any software engineer and better at software engineering than any statistician”. We may add to the whish list a deep knowledge of your businness domain and the ability to ask the right questions.

While not everybody agreed, it seems that the best way to start “doing Big Data” is one step at a time and with a project with clear bussiness goals. If you want to test the technology, good candidates are those business process that have already become a bottleneck using standard databases. On the other hand, innovation may also be an important driver, by using external open data or if you need to design data-centric products. A good example of that sort is the Open Innovation challenge from Centro de Innovacion BBVA,  providing aggregate information on  credit card transactions.

Textalytics

Finally, going back to the theme of our talk, one of the external sources that would is generating more value are social network data. Due to their heterogeneity, social networks are intrinsically difficult to analyze, but, fortunately, text analytics tools like Textalytics API, enable you to make sense of unstructured data. If implemented into your Big Data toolset they open the door to the intellingent integration of quantitative and qualitative data with all the valuable insights you would obtain.

If you want to dive into the Big Data world, videos of the talks and experts panel are available at the Big Data Spain site.

A %d blogueros les gusta esto: