Search Computing Blog

Yahoo! is promoting a very important initiative toward the democratization of research activities with big data. Such initiative, named Webscope, allows academic researchers to access a bunch of datasets, all of which  “reviewed to conform to Yahoo!’s data protection standards” on privacy.

Among the available datasets, great space is given to language and graph data, but some datasets also address important topics such as advertising, marketing and rating data.

More information about this initiative is available on the Webscope Website.

Researches from the Search Computing project attended the 2011 ACM SIGMOD Conference, which took place in Athens (Greece) on June 12-16.

A novel, live demonstration of the SeCo Execution Engine and Workbench environment has been presented at a dedicated booth.

DEMONSTRATION

Search Computing: Multi-domain Search on Ranked Data, authored by Alessandro Bozzon, Daniele Braga, Marco Brambilla, Stefano Ceri, Francesco Corcoglioniti, Piero Fraternali, Salvatore Vadacca

continue reading…

After organizing two workshops in Como, the Search Computing project decided to go “on the road”.  Several workshops have been successfully applied to conferences such as VLDB, ISWC, ICWE, and ECOWS. More details here, or on the workshops’ Websites.

At ICWE 2011 in Paphos, Crete (June) we organize the ExploreWeb
Workshop, chaired by Brambilla, Fraternali, and Schwabe, see:
http://exploreweb.search-computing.org/


- At VLDB, we organize the “Very Large Data Search” Workshop, chaired
by M. Brambilla, F. Casati S. Ceri, with Hector Garcia Molina
and Alon Halevy as keynotes, see: http://vlds2011.search-computing.net/

- Also at VLDB, we sponsor the DBRank workshop, chaired by Chackrabarti
and Martinenghi, Jan Chomicki is keynote speaker, see:
http://www.cs.uwaterloo.ca/conferences/dbrank/2011/

At ECOWS 2011 in Lugano, Switzerland (September) we organize the DATAVIEW
Workshop, chaired by Bozzon, Comai, and Norrie, see:
http://dataview.como.polimi.it

At ISWC 2011 in Bonn, Germany (October) we organize the OrdRing
Workshop, chaired by Della Valle, Horrocks and Bozzon, see:
http://ordring2011.search-computing.org/

In these times of  social networks and cloud-based productivity platforms, traditional search engines fail in providing tools and services able to collect and organize one’s Web information. Greplin tries to overcome such a limitation by providing a personal search engine for all that data you keep locked away in the cloud.

Greplin can interface with several Web social data service (typically, all those services that provide an API), like Gmail, Google Docs, Facebook, Twitter, and Dropbox. A full list of supported data sources is available here.

At the time, Greplin does not provide a public API to access its search functionality, but it looks like they are working on it.

Here is a presentation video of Greplin.

Greplin Demo from greplin on Vimeo.

In a recent blog post, Jim Jansen discusses about the outcomes of an expert study, funded by Microsoft, about the future of search.

The Future of Search

The study, which involved 54 search experts from U.S., EMEA and Asia, is available here; it contains the following major findings, all related with the Search Computing goals:

• Search will increasingly structure a wider range of data (social, UGC, video).
• Users want more than links to sites and documents with matching key words. They expect the engine to figure out what they really need.
• Aggregation of Data Silos: the search silo (index, history) will merge with other silos (social, location, purchase info, rich media, mobile apps) to give users what they want in fewer clicks.
  • For a movie, you might want to show info from different silos – expert reviews, friends comments, show times.
• Search Engines must extract more value from social data, possibly using semantic/natural language, and cultural inference to understand what people are really seeking.
• Search will be more and more ubiquitously enacted (PC and mobile).
• Community participation will play a greater role in search quality, using the social graph as filter (as, for instance, in Blekko), or using UGC to augment search results.
• Search engines will shift towards providing personalized results that leverage user history and context.
• Search engines will become more conversational as they go back and forth with users to refine a user’s query.
• Specialized vertical search sites are growing: users use vertical websites and apps to get faster, more relevant and more streamlined results. Mobile vertical apps are becoming increasingly popular because they reduce the number of clicks required.
• Real-time search will grow, but not everything requires freshness: explosion of real time information via Facebook and Twitter will drive user demand for fresh and new data, but everything does not need to be fresh to be actionable and interesting.
  • Nice example: MIT Sensable Cities project allows users to access real time data streams captured via cell phones as they move in urban spaces http://senseable.mit.edu/.
• There will be a growth of different UI paradigms for displaying information that is more relevant to the context of specific search query.
• The results experience will move from text and links to visually rich results, surfacing, for example, maps, weather charts/tables and other visualization of data as part of the result.
• Q&A will be important, expecially when obtained through community building.

We thank Alan Dix for suggesting the source of this blog post.

Google Shopping API

Google announced the release of the Shopping API, a new set of Web Application Programming Interfaces that are meant to substitute the existing Google Base APIs. The new Shopping Application Programming Interfaces (APIs) have two main components: Content and Search. Those components are part of a unique CRUD infrustructure for product data management.

On one hand, the Content API enables retailers to upload their product data to Google, and to make incremental updates to frequently changing attributes like price and availability.

On the other hand, the Search API provides access to product data. After creating a new project in the APIs console, a developer can issue JSON queries as the following one:

https://www.googleapis.com/shopping/search/v1/public/products?key=key&country=US&q=digital+camera&alt=atom

This query will return a feed pf products sold in the United States which are all matching the keywords digital and camera. With a registered account, the new Google Shopping API feature a default limit: 2,500 queries/day

The API supports both structured and free text search. Results can be ordered according to relevance, novelty, or price. It is possible to increase diversity in the set of products matching a query by using the APIs crowding mechanism to restrict the number of products with an equivalent property.

The Google Base API will be fully deactivated on June 1, 2011. Some non-shopping data types (such as jobs, real estate, events, and activities) won’t be supported anymore.

Yebol is a meteaseach engine that utilizes a combination of algorithms paired with human knowledge to build a Web directory for each query and each user. Instead of the common “listing” of Web search queries, Yebol automatically clusters and categorizes search terms, Web sites, pages and contents.

Yebol claims to allow for multi-dimensional search results, but the actual meaning is that it can explore at once several web sources (web, images, twitter, videos, and so on). So: is it just a metasearch engine with a nice interface or something more?

As many other Web 2.0 applications Mendeley also has Web API. Check out Read Meter by Dario Taraborelli for a great mash-up built on them.

Search engines are exploiting more and more named entity identification in the query analysis phase.
Besides increasing the precision of the results, this enables the generation of result pages more suitable to the typical needs of users with respect to the identified entities.

The attached document considers queries returning mono-domain results, where the domain represents a specific field of interests such as City, People, Movies, etc. and characterizes the problem for the definition of the layout of such results. In particular, it analyzes the behaviour of the main current search engines (Google, Bing, Yahoo) according to the result page layout definition issue.

search engines: page layout analysis

The final objective is to describe a conceptual definition of the Web search result layout problem, by identifying: the parameters involved in the layout design, the tuning dimensions available for optimizing the result layout, and the possible strategies that can be adopted for producing such layouts.

This has been the topic of a position paper at the DataView workshop within the OTM conference 2010, Crete, Grece.

While most of the people identify Web search with Google, experts of the field and Web addict know that there is much more than that around.
New search engines are coming out every day. Here are a few examples that somehow relate to Search Computing:

1. Goby (http://www.goby.com/)

The payoff tells you all: “Create your own adventure”. Pretty similar application to our “plan a night out” scenario and demo. Several kinds of results are returned, categorized and associated with several additional details (photos, address, purchase info). I see three big differences with respect to SeCo:

• no clear and configurable ranking criteria are exposed
• only items of one single type are returned (e.g., a list of concerts)
• no exploration towards other item types is allowed

Goby is available also for mobile:

2. FanSnap (http://www.fansnap.com)

This is a vertical search engine for celebrities and events. It’s focused on ticket purchase. Again, just one concept at time (in this case, the event) with fixed sorting criterion (the date of the event).

3. SurfCanyon (http://surfcanyon.com)

SurfCanyon aims at improving the understanding of user intents by asking for suggestions while the search is performed.  While the first searches retrieve diversified results to try to catch all the intents, after some executions (and explicit declarations of the user) the system learns the typical user intents and exploits them in the subsequent searches .

4. Blekko (http://www.blekko.com)

This one appears as a classical keyword based search engine, but it provides a new feature: slashing the web. This means that search keywords can be combined with slash commands (e.g., /liberal, …) that tell the engine to retrieve only contents that contain opinions in line with the viewpoint or political orientation specified by the slash.

5. Siri (http://siri.com/)

This engine is a good representative of the specific category of question answering / task solver systems. It lets the user state his need and tries to accommodate it. Categories and concepts are recognized and retrieved. Context from previous searches is also considered.

6.  Mobile search engines

Here is a shortlist of the search engines now available also for mobile platforms: