AUTHOR
Miranda Mowbray
ABSTRACT
This paper discusses an issue that may arise from the combination of the accelerating growth in the number of pervasive computing devices, and the continued rise of social computing, in particular microblogging. The issue is unwanted friend requests from inanimate objects.
A Twitter user who wants to be sent in real time the tweets published by another Twitter user’s account, or to read private messages sent by that account, must first send a friend request to the account. The act of sending a friend request to a user’s account is also known as “adding” the user. A tweet sent by Eugene Huo said:
Man, I gotta watch what I talk about on twitter… as soon as you mention something they find you… a rice cooker just added me.
It was probably a company advertising rice cookers rather than an actual rice cooker that added Eugene Huo, but in the future it might well be a rice cooker. Pervasive computing devices are already using Twitter as a communication channel, in growing numbers. Everyday objects that tweet include a toaster (@mytoaster), a garage door (@connectedthings), shoes (@ramblershoes), ovens (@bakertweet), a house (@ckhome), and a catflap (@GusAndPenny). There is a home security system that sends public tweets about which doors are open – this is perhaps not a good idea.
Most of these objects send out data on Twitter but do not receive it. However, Twitter offers an easy-to-use channel through which pervasive computing devices can receive data as well as transmit it. Pervasive computing devices on Twitter may be able to gather useful information from other devices and from human tweeters, including tweeters who are strangers to the owner of the device. The potential of such information is demonstrated by the fact that measures based on human Twitter activity have been found to predict the result of the most recent UK general election and box-office takings for Hollywood films better than the previous best prediction methods. Information from human tweeters can form a rapid and flexible service in emergencies. Company service representatives are already using Twitter by following customers and other users who tweet words connecting to their brands, and responding to any complaints: in the future these companies’ products themselves might use Twitter or other social networks to automatically mine real-time information from customers that could be used to enhance the service provided by the product, or proactively solve problems.
Moreover, the Twitter interface offers an intuitive way for human users (especially remote users) to interact with a service. A playful example of this is @tweet_tree, a Christmas tree whose lights change colour according to commands sent to it by Twitter users. Other automated Twitter accounts have looked up stock prices, announced the football results for a user’s favourite team, or played games with users. In these cases the user controls the automated account by sending it “direct messages”, which are private one-to-one messages that can be sent by a Twitter user to any of her followers.
This raises the spectre of human users being bombarded with friend requests from inanimate objects with embedded computing systems, where the owners or manufacturers of these objects wish the objects to receive potentially useful information from the human users in real time, or wish to enable and encourage the users to interact with the objects via direct messages. In fact unwanted friend requests are already a problem on Twitter, because the most common Twitter spamming technique involves automatically sending friend requests to a very large number of users. A user who accepts and reciprocates such a request may find herself receiving unwanted communications from an automated account – whether this is the Twitter account of an inanimate object, or a spammer, or even both at once. Or her private tweets may be automatically harvested and used in ways that she does not wish. But even if she refuses every one of these requests, receiving very frequent friend requests can be annoying in itself. It may also lead to the user mistakenly refusing requests from some people (or objects) that she would in fact like to be friends with on Twitter.
To address this, some form of validation service is needed to help users decide whether a requesting account is likely to be one that they wish to communicate with, and to enable them to automatically refuse or accept requests from some classes of accounts. Several Twitter validation services, most of which have the aim of identifying spamming accounts, are currently available. Unfortunately they have some limitations, the most common being that their accuracy could be undermined by changes in popular software for automated Twitter accounts. This limitation is shared by some measures for spammer identification suggested in papers by Lee et al. and Stringhini et al. Indeed, several of the measures used by account validation services are unable to distinguish benign Twitter accounts from automated accounts that operate some types of spamming software in current use. One particularly interesting service, TrueTwit Basic, uses a CAPTCHA challenge to attempt to discover whether or not an account is automated, under the assumption that human users do not generally wish to grant a friend request from an automated Twitter account. But this assumption may become decreasingly valid as useful pervasive computing services move to Twitter.
In addition to pointing out limitations, this paper identifies some useful features of several validation services that might contribute to a more robust solution. If we cannot resolve this issue we may find ourselves befriended by hundreds of rice cookers.