
InterMix has messaging and cataloging capabilities well beyond anything currently available. We think these will give it the marketplace appeal it needs if we are to succeed in our grand venture.
Delphi and Java volunteers are wanted to code thick and thin clients while we concentrate on the InterMix internals and the Recommender form generator. Internet savvy volunteers, don't stop to think :-), read on, and then introduce yourself to rogereaton@earthlink.net. We need help at several levels, first with organizing the effort, then with developing specs, coding and checking. Latest functional specs for InterMix are available as a 130k .wri file here.
Any number of input fields may be specified, each with its own parameters: length of field, number of entries, whether the field is a required entry or not, optional template or verification from a given list, and so forth.
Sorting. Up to three custom sort orders may be specified for a given Recommender Form. A sort specification may include data from more than one field. Using movies as an example, one field might have the year of the film, and another the name. Then one sort order might be by year and name. In addition, all items may be sorted by the date and time they were entered into the database, by average interest or approval rating, by cumulative value rating, and by controversy. See below for details about Ratings and Sorts.
Selection. Any field may be indexed, and when a field is indexed, it is possible to select by the value of that field. In the movie guide example, the "actors" list would certainly be indexed, so one could easily select, say, all movies with Sean Connery, and then have them sorted by year and title. In addition to the customized indexes, InterMix provides a number of standard selection capabilities, including selection by keyword, selection by database area, selection by rating, and selection by the date the information was entered into the database. Users known to the InterMix hub may also specify that only items never browsed before be listed.
Hypertext Links. Any Recommender field may be linked to a separate Recommender form. A pop music database, for example, might consist of several Recommender forms: for the song, for the recording, for the artists and writers, and for the package. The page for a song would include a list of known performances. Clicking on a particular performance would give more detail -- the year, the complete list of artists, the albums and cds and tapes where this performance has been published and so forth. Click on the artist to see a bio and list of credits. Click on a credit to see another performance, etc, click, click, click!
Automatic Cross-reference. Two indexed fields in separate Recommender forms may be tied together so if information is entered on one form it is automatically updated on the other. For instance, a movie database with one form for movies and a second form for actors, could be set up so that adding an actor to a movie would automatically add the movie to the list of credits for the actor.
Extensions. A special kind of Recommender form called an "extension" may be included in another form at the time data is entered into the form. For instance, in a database of community organizations, a "non-profit" extension might include fields for mission statement and IRS status. The basic form would include name of organization, address, contact and so forth. The non-profit fields would be included only when needed. In this way a single database can include many sub-varieties without cluttering up the basic form with specialized fields.
Extended Arrays. Named extensions can be integrated into the basic Recommender form as a method of creating multiple field arrays. For instance, an array of awards for movie actors might include a three field extension for the name of the award, the year of the award and the associated movie.
Exportable. Recommender text can be exported as a field delimited file.
The InterMix Engine itself is based primarily on the "item" object. InterMix areas, users, messages and Recommender forms are all "items". This means they share a number of useful properties, most importantly, the ability to be selected by keyword. The result of this design is that InterMix is quite flexible internally. New features can often be added with relatively little trouble.
At the core of the Engine is a special keyword index file which enables selection by keyword in conjunction with a number of built-in or customizable sort orders. The work is done by building indices when the user adds an item to the database. Switching from one sort order to another requires only a change in lookup index, no actual sorting. This combination of indexing and sorting is essential for implementing collective communication, and it makes possible InterMix auto-session capability. It also is the basis for our Collaborative Catalog strategy, which will allow InterMix providers to combine their databases in robust Web sub-nets that can tolerate the continual coming and going of sub-net partners.
Collaborative
Catalog Strategy
Imagine a linked database spread across twenty internet servers that are subscribed to a subnet. If we want to select and sort items from the subnet into a single list, it's not so easy. It looks like either we select first and load the selected items from all twenty servers into one of the servers and then sort, or we have each separate server select the items and then sort them into temporary databases for us to query one at a time from a central location -- an expensive operation either way. But the InterMix presort by index is a third solution, allowing us to retrieve the highest order item from each of the twenty servers efficiently. The retrieved items are compared and that one which is first in the sort order is put on the output list. Then a request is sent to the server which provided the high order item for a second item. Again the comparison is made and so forth until the output page is complete.
Essentially this method is a tradeoff of flexibility for speed and amount of work required. Only sort orders which have been preestablished for this particular Recommender item type are supported, so InterMix is not as flexible as some method that more tightly integrates units across an internal network (intranet). The InterMix approach, though, is more practical for the rough and tumble of the internet.
Collaborative catalogs will be implemented as a detail within the more general publish and subscribe EchoNet system. See directly below.
Publish
and Subscribe EchoNets
Our expectation is that the web will become more and more a two-way medium, with ever greater numbers of only moderately sophisticated users setting up web servers and web sites on their own home computers. Our targeted customer is precisely this sort of internet hobbyist who wants to participate in either a collective communication network or in building a collaborative catalog. The thick client is a natural introduction to the inner workings of an InterMix hub, a confidence builder on the path towards full web participation.
The approval ratings are a form of online vote. The -3 to +3 scale has both negative and positive possibilities so the voter can express basic for and against feelings, and the gradations allow a second best or third best choice in case the voter's favorite fails to win group approval. Ratings are better than simple yes or no voting at handling priorities.
The interest ratings run from BOMB! to **** like movie ratings. How much does the rater want to see another item like this one? That is what the interest ratings are supposed to measure. But why should anyone part with this personal information? And more subtly, won't the interest ratings be used as a kind of vote? InterMix items can be listed in order by average interest rating, so the higher the rating, the more likely others will view the item/message. Giving a high interest rating thus promotes an item or message.
For version 1.0 of InterMix, there is little we can do about these problems. Further down the road, though, InterMix will profile the users from their interest ratings so it can steer likely items their way taking individual taste into account. This will give users a reason to provide true personal ratings. (Customized InterMix shopping agents will be the commercial draw that powers our Humanity 2001 project to its goal.)
For Recommender items approval ratings may not make sense. A book Recommender designed for fundamentalists might indeed have approval ratings(!), but for the rest of us, books can be handled on a single scale of zero to four stars. The Recommender is flexible, allowing either, both or neither of the scales to be implemented for an item type. Also, the name and description of the rating scale can be changed. One idea we are considering is to use the approval scale to rate the quality of item information. For instance, in a movie catalog, if Citizen Kane were listed as a 1955 movie, that item would deserve a -3 for "info quality". We can use this kind of rating in our collaborative catalog strategy for choosing between near duplicates from different sources. Open collaboration needs a method like this to clean out the mis- and dis-information. (Easier said than done, though.)
The value of InterMix rating sorts is greatly increased by our Inclusive Area Hierarchy design. InterMix areas are hierarchical, like Usenet newsgroups, but unlike Usenet newsgroups, the higher levels of InterMix areas can be accessed, and they include all the messages in the sub-areas. It is as if you could specify the newsgroup ALT and see all the messages in all the ALT newsgroups in one list. Not a very useful exercise, you may think, but the picture changes when you realize the messages can be listed with the most interesting or controversial ones at the top.
Sort by date. This is a standard sort, often useful in reverse order, most recent item at the top of the list. When items are imported from another hub, there is a choice whether to use the original posting date or the import date for date sorting.
Sort by interest. This is another standard sort, available for all InterMix items and messages. The sort is by average interest rating. Items with no ratings are listed as having exactly average rating, and items with just a few ratings are regressed to the mean. It is possible to request a list of all items on an InterMix hub in descending interest order.
Sort by approval. Just like the interest sort, but based on average approval rating instead of average interest rating.
Sort by controversy. Another standard sort. Items are listed by standard deviation of the approval ratings. Standard deviation is a statistical measure of the "spread" of a set of scores. The more spread out the scores are, the greater the standard deviation. For InterMix approval ratings, the highest standard deviation would be generated if exactly half the raters gave a -3 and half a +3 approval rating.
Sort by value. The "value" of an item is calculated as the sum total of all interest ratings times the average approval rating. If the average approval rating is negative, so is the calculated value. An older item that has stacked up very many interest ratings, even if they are on the low side, may show greater value than a new item with a high average, but only a few ratings. Topical items whose interest fades with time will not accumulate much value.
Custom sorting. See The Recommender section. Up to three custom sorts are available for each Recommender item type.
The area hierarchy is "inclusive" because messages posted to sub-areas are included in all the super-areas up the line. A message posted under BASEBALL.REDSOX is visible there with other Redsox messages. The same message is also visible with messages about other baseball teams under the BASEBALL area, and at the top, all SPORTS hub messages are visible together in one very long list.
All the messages in one loooong list! What is the point, you may ask? The whole purpose of splitting areas into sub-areas on the internet has been to keep the lists manageable -- long lists have been something to be avoided. But now, the InterMix ability to sort by interest and by controversy changes everything.
When the interesting messages can be floated to the top, it doesn't matter so much how long the list is. This takes some of the pressure off within an area, but more importantly, it adds a new dimension to the entire message exchange process. "Interesting" messages are now doubly interesting. First for themselves, but then, too, because these are the messages that everyone will see. The tendency of the internet to split people into tiny sub-specialties will be balanced by a sharing of messages at every superlevel, right on up to the top where the most interesting or controversial messages from anywhere on the net will come to general attention.
For instance,
Hubs may be private or public, and public hubs may either allow or disallow anonymous browsers. A private hub always requires a separate login, and private hub information is never automatically exported to another hub.
Multiple hubs may be implemented on a single web server and served by a single InterMix host. At the option of the sysop, these hubs can know about each other, and users can easily switch from one public hub to another without logging in again.
Each hub is a separate unit. Until we implement our Collaborative Catalog strategy, and our Publish and Subscribe networks an InterMix item list will contain items from a single hub only. Collective communication is also based on the hub, which is the smallest unit that can generate a collective message.
File upload as well as download is available for those with browsers that are capable of uploading. If browser upload capability remains flaky, as it has been, we will look for another option, possibly email or an ActiveX applet.
Chat will be implemented in Java.
Automatic text keywording and search is already available. Every word of text is indexed except for the 100 most common English words.
Automatic hot buttons make it easy for the user to find active areas and threads.
An email inbox is provided each registered user. This is intra-hub email only, not internet email. An "outbox" and work-area keeps track of all messages written by the user for the user's reference. Messages that have not yet been read by the recipient can be deleted through the outbox facility.
Pseudo archive ability allows the sysop to automatically "archive" old messages, thus removing them from browse lists unless specifically requested.
The html "direct display" function enables direct display from an InterMix hub. For instance, the sysop might set up a direct display button for Los Angeles business contacts. This button would use CGI to display an html page listing all L.A. contacts from an InterMix hub. Users would not have to log in to the hub to display the list. The sysop would have convenient remote control of list contents through the "Los Angeles Contacts" area of the hub.
The "Censorship Control Panel" gives the sysop four levels of censorship control. Level One is unbreakable. Only the sysop can see items censored at this level. For levels 2 - 4, users can control censorship for themselves, turning it on or off at will level by level. For instance, uncivil messages might be censored at level two, and pornography at level 3, so users have their choice of leaving porn on and turning flames off or whatever.
| drawing "Miranda" © 1996 by Sandy Chaves |
InterMix is also known as XimRetni. Last changed July 1, 1998.