OpenCOLA

OpenCOLA, a Toronto-based company with a unique marketing strategy, just released an autonomous and collaborative open agent framework under the same name - where COLA stands for Collaborative Object Lookup Architecture. It is a true collaborative computing environment, in which all instances of the program are peers. To avoid referring to "clients" and "servers" in such a peer to peer environment, the OpenCola platforms are called "clervers". Another peer-to-peer startup trying to promote yet another way to share electronic documents and MP3 files? When combined with another key concept, expert communities, it could actually offer an elegant solution to the problem of overcrowded indexes of today's search engines. You will share your hard drives with all other users who have similar interests. Your personal agent (stored as XML file) will be trained to recognize and seek out only the specific information. The whole idea relies on the network of users with equal rights, essentially spidering and indexing small chunks of the Web in their spare time.

Several applications should use this concept for achieving different tasks. OpenCola Swarmcast will be a large file distribution system, allowing content providers to serve large files significantly faster, breaking large files (> 1 MB) into multiple, easily transferred chunks. These chunks will be served simultaneously from multiple nodes in the OpenCola Swarmcast network. OpenCola Streamfind is an extensible tool for announcing, distributing, searching, filtering and viewing streamed media content. However, the most interesting products from the AI standpoint will be called OpenCola Smart Folders. Using a combination of machine intelligence, human decisions gathered from across the OpenCola network, and direct user feedback to improve results it will continuously and transparently search and retrieve the interesting information: The way OpenCola Smart Folders works is simple: Name a folder. Drag a file into that folder. The folder automatically fills up with other files that are similar. If you take something out of the folder, it doesn't find stuff like that anymore. You can create as many folders as you want by setting them up yourself or by adopting them from other users.

Early adopters and experimenters can now download the OpenCOLA SDK. Documentation is still incomplete, so the knowledge of C/C++, XML and HTTP is absolutely essential.

WDBC

LOTONtech Limited recently released Web DataBase Connectivity, a Java2 adapter that lets you run SQL queries against live Web pages as though you were querying relational database tables using JDBC. It uses an enhanced version of SQL - called HTMSQL - that allows hierarchical HTML data to be accessed according to the flat two-dimensional relational database paradigm. I already dedicated an earlier feature article to the similar tools for "SQL-style information retrieval". WDBC shares a similar approach, and will be very attractive for Java programmers building search engines, data mining tools, or knowledge-based applications. Beta release applet can be tried from their Web pages. There are still some bugs waiting to be fixed, but it offers an intuitive and easy way to extract information from the semi-structured Web data.

DolphinSearch

DolphinSearch, Inc. was founded in 1999 to exploit Dr. Herbert L. Roitblat's pioneering research into how dolphin brains process echolocation signals. The research has led to breakthroughs in using pattern recognition techniques to enable machines to recognize the meaning of human language. They offer a family of technologies for automatically reading documents and storing them in a searchable database that is very different from the search engines as they are used today. The main concept behind this technology relates relational pattern recognition mechanisms that model how the dolphin recognizes objects to the text recognition techniques. "Just as a sound wave changes its meaning depending on the context of other waves around it, so too do words change their meaning - words have senses, nuances, and multiple meanings - depending on the context of other words around them."

According to Dr. Roitblat, "knowledge management should be an inherent network function". Following that principle, their main product, KnowledgeBox, is a plug and play, hardware-based network appliance capable of finding any unstructured information stored anywhere on a computer network. KnowledgeBox quickly scans through the text on a company's network, regardless of format, understands its meaning and remembers where it is stored. Pricing is $10/seat/month, with a minimum commitment of $10,000 (US) in the first year. In subsequent years the minimum annual commitment drops to $2000.