Archive: Press coverage - Automating Moderation Services::
(Originally published 18 September 2003)
Source: UN, 5 February 2003 Submitted by Ann Light
Moderation is a specialist, but growing and costly, service that many companies employ to oversee online chat, discussion and other postings to their sites. Its profile has recently risen with the publication of Home Office guidance on protecting children online (see UN story Government Report into Best Practice for Children Online). However, even without this glare of attention, the value of moderation to avoid legal tangles, protect brand image and encourage user involvement is well established.
Emint, a group of strategists and managers in charge of moderation services and software, met last week to discuss which aspects of online moderation can be automated and what associated complications this might create.
Speakers were Tamara Littleton, founder of eModeration.com; Sara Tebelius, communications manager for chatmoderators.com, and Adrian Barrett, founder of lightmaker.com specialising in 3D virtual communities. Littleton stressed that the job as a whole cannot be automated, but that much of the administrative/management part of it might be. She and Tebelius drew attention to the subjective nature of many interventions made by moderators. While software can flag up postings containing specific words that have been deemed unacceptable on a particular site, it cannot deal with the subtleties of ethos or allusion.
Allusions could be especially treacherous. One person gave the example of footballers involved in court cases - something which happens all the time - being talked about on a football site. Discussion of anyone's involvement in a court case mostly breaches the law of contempt of court, with potentially severe penalties for the publisher. A filter can pick up names that shouldn't be mentioned at such a time, but not oblique references such as the footballers who.... And no one wants a filter that raises every instance of footballer for extra scrutiny.
Some of the discussion thus echoed that heard when natural language processing is discussed in other contexts, such as search: how to produce software that can arrive at a person's meaning from what they say.
It was also pointed out that what was offensive was also related to brand. Which words did the Emint group itself find unacceptable? "Bollocks"? One speaker described the novel and automatable solution of substituting offending words with random nouns, making for a bizarre and sometimes funny posting that softened the impact of the editing. This light-hearted style went down well with the particular users and might work elsewhere.
Different tactics for dealing with removing or modifying postings were also raised. In some contexts, public removal of inappropriate material - combined with a note documenting why - is necessary to make a point. In others, everything is handled behind the scenes with a word to the poster. This aspect of relationship management is crucial and also cannot be automated as a whole.
Barrett raised the temperature of discussions by showing his company’s 3D avatar-based system. The 3D world turned the private chat-room of the text-based discussion forum - now regarded with increasing suspicion as a place for snaring children - into a full-blown apartment, with fire in the corner and rug. Other seductive features also struck the group as making moderation more difficult. Barrett commented that companies were resistant to the cost of moderation on top of the cost of installing systems of this kind. New challenges were in store.
Lizzie Jackson, head of communities at the BBC, who convenes Emint, ended the session with a list of areas where good design can support the moderator in their current tasks. Those related to software included:
- an alert button for users to push if unhappy with another's posting,
- safety bots, that promote safe behaviour within and beyond the forum,
- statistical signposting, where the best messages of the day are displayed as a role model for other postings,
- profanity filters that can be set to flag up or replace words that fall outside what is acceptable for particular services.