We are happy and proud to announce that the Vietsch Foundation will support the further development of Searsia. The Vietsch Foundation was founded by Willem Karel Vietsch (1952-2014) through his last will and testament:
Karel was very involved in the Internet community. He spent a large part of his life working in the area of research and education networking, which he help developing with his energy, creativity and thorough methodical approach. Sadly, his early death, put a stop to Karel’s mission to support people and new developments, therefore he created the Vietsch Foundation, with the official objective of promoting “research into and development of advanced technology for scientific research and higher education”, in the widest sense.
Searsia is funded from the “Research and Higher Education Technology Fund”, a new thematic fund that is managed by the NLnet Foundation. The Vietsch Foundation will fund the Searsia work packages: Recommendations and Advertisements.
Recommendations without tracking of users
Searsia manages large collections of independent sources. Many advanced features of search engines might be seen as sources too, such as query autocompletion, related queries, and spelling correction, that is, these features are provided by modules that receive the query, and return a recommendation. Query autocompletion provides users with suggested queries in a drop down box as they type their query. Related queries are given on the search page to direct users to alternative formulations of their query, or to otherwise related information. Spelling correction suggests corrected queries if the user types an incorrectly spelled query, or an otherwise unusual query. Corrections are either used directly or presented to the user as alternatives (“Did you mean: …”).
There are two main approaches to recommendations, often used in for instance movie recommender systems: 1) collaborative filtering provides recommendations based on actions from similar users (as in Amazon: “Customers who bought this item also bought…”); and 2) content-based filtering provides recommendations based on the similarity between items. Autocompletions and related queries in search engines are almost always based on queries from other users, i.e., they are essentially collaborative filtering approaches. Besides the fact that it might be undesirable to log the actions of the users, there are also other drawbacks: First, there is a cold-start problem: How to implement recommendations for a new system (when there are no user actions yet), or for a system that does not have a lot of users? Second, user queries might be stupid, racist, or otherwise objectionable: How to prevent recommending those? (it is not hard to find examples on-line of search engine autocomplete fails, as well as law suits about objectionable recommendations).
We will implement and demonstrate open source solutions for recommended queries that use content-based filtering, for instance based on a web crawl, or directly based on the Searsia sample index (The Searsia server indexes search results for each resource to learn what information each resource provides). Our recommendations work without user tracking.
Advertisements without tracking of users
Advertisements are the main source of income for many web sites and search engines. To make Searsia an interesting solution for developers of web search applications, we want to provide them the means to easily and transparantly add advertisements as an independent source to Searsia. The advertisements must be recognisable as advertisements, they should not track the users of the search service, and it should be easy to switch them off. Advertisements in search engines are much more effective than advertisements on ordinary web sites, because users tell the search engine exactly what they are currently looking for: The user’s query is the ideal signal for an advertisement network. We believe user tracking is not desirable and not needed for good advertisements in search applications. Searsia technology should (also) work without user tracking.
Searsia uses a broker (the Searsia server) that acts as an intermediary between the sources and the user. We will implement advertisement such that the user’s queries will not go directly to the advertisement network. Instead, they are forwarded by the broker. So, the advertisement network will not receive information that might identify the user, such as IP numbers, cookies, and the browser’s user agent string, which makes it virtually impossible to identify or track the user.
We will find at least one, and likely more advertisement partners that share our vision, and that are willing to provide advertisements for anonymous queries from Searsia search engines.
More information
For more information, check out the information at the Vietsch Foundation and NLnet.