ETS Technical Draft
The CouchSurfing Wiki, an informal workspace which anyone can edit.
Technical Draft for the Emotional Tracking System
Contents |
INTRODUCTION: Building some foundations of the ETS
Through deconstructing the emotional tracking idea, let's see if some of the components or steps to its creation can be easily integrated into the current CS system. Perhaps some aspects of this approach can simultaneously address other issues of concern to many in the CS community.
A necessary foundation of an eventual emotional tracking system, and what we might call 'self-evolution support', is friend-making or friend-matching. Facilitated friend-making, which we can do through both software and the harnessing of member activity, is a natural evolution of the social network paradigm, as well as something that can extend primary goals of the couchsurfing ethic while serving the membership in greater ways than other similar travel websites.
In turn, a necessary precursor to friend-making is a strong basis of trust. Given the online nature of CS and its complex intersection with humans in the real world, the primary bases for trust reside in CS features like profiles, friendships, and vouching, as well as 'in-person' aspects of hosting, surfing, and meetups. However it should be strongly noted that trust must exist BEFORE the surfing or hosting occurs, and that lack of trust is the major impediment to a larger number of people participating around the world. Thus, while the current CS framework has served admirably well, it is crucial to ensure that the system itself scales its trustworthiness with the rapid increase in CS membership. CS is not merely an online community, but a network of travelers who to some degree entrust their safety to other CS members in hosting and surfing. Therefore, making the system as trustworthy as possible and reassuring members as to that trustworthiness - while winnowing out members of bad intent and attempts at social network hacking - is of paramount importance.
In designing a novel ET and friend-making system, we can also multiplex some of its features to increase security through trustworthiness. In this document we address two such components, -'facilitated friend-making (FFM)' and 'trustiness'.
FRIEND-MAKING, Friend-matching, Friend-building
One of the two main components of the Emotional Tracking System tem], Facilitated Friend-making (FFM?) or -matching is in many ways a basic goal of Couchsurfing. Leonardo made some excellent points in the original wiki about the difficulty of finding compatible friends, how much we can learn from them, and how valuable a system that assists everyone in making friends (and eventually, connections that we can explicitly learn from) would be. Problems here lie primarily with the 'match-making' aspect being seen as a dating site, and the resulting kind of people that would be drawn to that instead of to the primary cultural-transmission and learning goals of courchsurfing.
There are multiple levels of 'friend-matching' or 'facilitated friend-making' that we could implement:
1.-human-driven: a fairly easy extension of the current infrastructure, a member-driven friend-matching system could present simple options like 'connect two friends or people' or 'suggest shared travel'. This would allow members to note which of their friends they think would like each other, and promote a more formal connection of those friends through a generated email suggesting chat, email, or a meetup. Right now this is often done informally through email on or off the system, but a more formal method would allow CS to begin tracking how often suggestions result in friendships or other connections, extract what shared characteristics might have contributed to the new friendships, or determine 'expert system' type data on how people who are very successful at friend-matching, are able to do so. People who are consistently good at facilitating friendships could also have that integrated into their "trustiness" measure.
2.-system-assisted: implement some keyworded (and additional) features similar to those seen on Facebook and other social networks, for instance making all movies, music, etc noted by a user, hyperlinked to the group of other people who also noted them. This allows members to more easily see those with shared interests, perhaps generating more compatible surfingss, and spawning groups, meetups, or chatrooms based on those shared interests. This could work off the 'tagging' , 'tagclouds' and 'tag overlap' that Thomas mentions.
3.-system-derived (automated) : this is a far more complex endeavor, as it will require fairly extensive questionnaires or explicit choices, additional extracted data, and probably an 'opt-in' approach. The idea here would be that the system itself would generate suggestions for friends, sending an email notifying both parties when some (currently undefined) threshold is reached.
All of the above would incorporate, to some degree or another, the 'trustiness' factors described in the section below. Some additional features or measures could also be incorporated (such as direct ranking of profile 'authenticity' by members who have met each other) that would be especially helpful in the 'system-assisted or -derived' versions.
This overall 'facilitated friend-making' approach could be extended into a variety of areas, such as linking those who will be traveling to the same places at the same time and suggesting meetups or shared travel, noting when events (such as paragliding meetings) are taking place and alerting groups about them, or otherwise pointing people toward those likely to share interests, activities, or modes.
"Friend-building" and possibly 'friend-keeping' belongs to more central aspects of the emotional tracking system not yet addressed here.
"TRUST RATINGS" - Building a superlative trust network
Trustworthiness has a fundamental impact on the ability to make friends, keep friends, deepen friendships, and extend connections throughout the global network. This is especially true when connections are made principally with strangers.
Some of the ways trustworthiness impacts the ability to make friends on CS, are:
- Our direct interactions with individuals;
- Our perception of others' trustworthiness, given by our impression of what they have written in their profiles;
- The rankings and assessments in a profile, i.e. friendships (quantity, quality, descriptions), references, verification and vouchings;
- Our ability to assess the trustworthiness of others who have made assessments of others (including references, friendships, and trust rankings);
- how our own trustworthiness is perceived by others, including all of the above as well as possibly accounting for perceptual and cultural mismatches and miscommunication.
A potential problem with the current CS system (as it becomes better known and a target) is that there is no measure of overall trustworthiness of individuals taken into account in the system. This means that it is vulnerable to social-network hacking, wherein people can create false profiles, give false references in agreement with one another, and otherwise propagate false information and give false impressions. While checks and balances exist within the current system to counter these, there is no active 'winnowing' of such false information or people, and it can be somewhat difficult to determine the trustworthiness of people who do not have a great deal of activity in the system.
One solution is to strengthen many aspects of the trust network. One approach to this is to create a kind of overall 'trustiness' rating that reflects differences in proven trustworthiness - between for instance, CS Founders, and newly arrived participants, between those with good references vs bad ones, and so on. While trustworthiness is implicit (and in some cases explicit) in many aspects of the system, a large amount of the information currently embedded in the system is not transparent to members, especially new members who might have most need of it.
A measure of trustworthiness could be be taken into account when determining the validity and weighting of references and other assessments. Currently references by new members are presented by the system as being equally valuable as those given by longer-term CSers with extensive experience, a CS ethic, and no personal agenda. Introducing an additional trusworthiness measure could be useful, for instance, in addressing the problem of new or malicious members (perhaps with a non-positive agenda, who believe this is a dating site, etc) who leave negative references (a recent case in India). Since the problem of spurious or retaliatory negative references inhibits members from leaving negative references (because negative references usually result in a negative reference 'payback'), weighting references by the trustiness of *who* has left it (as well as their actual ranking of negative-neutral-positive) provides a more accurate indication of the value of a reference.
Example: a negative reference is left on the profile of a person you are deciding whether or not to surf with. This leaves a negative impression. To check it, you can then go look up the person who left the negative reference, and determine from THEIR references if they are to be trusted or not, by the friends, references and so on that others have left them. But are THOSE people trustworthy? If the person who left the negative reference is a new member with no surfing experience but many friends, all at level 3 and met through chat, that would implicitly indicate one degree of trust in their negative reference. If the person who left the negative reference is a Founder, a highly trusted individual with concomitant friendships, references, vouching, etc, it would carry a far different *implicit* weight.
Right now extracting this implicit data takes a fair amount of work AND knowledge and is not likely to be taken into consideration all the time. While usually this still results in a safe experience, it allows cracks in the system to be exploited by ill-intentioned people. (and yes some of us are paranoid about this as the network grows and the message and intention can be lost on new people with their own agendas) A newbie for instance, would be unlikely to be able to navigate these parameters or integrate the information very well, despite the fact that they would need it the most.
To build a stronger trust network we suggest an extension and expansion of the current system, through a recursive use of 'trust ratings'. A trust rating attached to each member would allow a deepening of the trust network, and a propagation of trustworthiness ratings throughout the membership network.
Trust Rating
A trust rating is a summation or integration of a multitude of weighted factors that together yield the overall trustworthiness, or 'trustiness' of a given member. By extension, this implies how trustworthy their references, friendships ratings and so on, are. One way to envision the trustiness of a member is as the equivalent of Ebay's overall approval rating – except better ;).
The way we could use this is to have a percentage (92.8%, 3%) or set of 5 stars (filled or not) that goes next to each member's name, whether on their own page or next to their names as friends, giving references, etc.
All trust ratings would be scaled such that a Founder's (or other trusted person's ) positive reference is factored in more heavily than a negative reference by a new person who hasn't even filled out their profile. While to some extent 'vouching' does this, it applies to only a small subset of the members (and possibly an increasingly small subset of members if membership growth rate increases), and does not provide data on the great majority of members. Verification, on the other hand, provides an alternative, but has its own drawbacks.
The algorithm used to determine the trust rating would be based on the integration process used by neurons, which integrate a set of varied, and often dynamic, weighted inputs. For couchsurfing, this approach would incorporate factors such as duration and frequency of participation, vouchings and verifications, and ratings given by others (friendships, references, etc). Most of these factors are already explicitly accessible within the CS framework and thus easily available. A few other factors whose primary purpose would be to enhance friend-making or be used in the ETS, could also be usefully incorporated. For example, ratings of a person's profile either by the system itself (rating decreased by empty sections, penalizing people for a poorly-filled profile), or by members who have met the person, could be used.
Some (not all) factors that can be integrated for an overall trust rating:
0. Spam reports, or other complaints.
1. Vouching - clearly this should carry a great deal of weight as it is essentially a measure of trusted people explicitly saying that others are trusted.
2. Verification
3. Friendships:
- Number of friendships x strength of friendships : (weighted so that quantity is weighted less than quality, but to be determined)
- Duration of friendships: longer friendships are more strongly weighted, but some thought should be given as to whether 'outside' friendships brought into the CS community should, given other factors, also be strongly weighted.
- Direction of friendships – track the changing ranking given to the friendship between two members and note its directionality: have they grown to be closer and deeper friends, or has a good friendship become less so? While changes in friendship are somewhat normal and thus should not be too strongly weighted, the weighting of decreases in friendship ranking or deletions of friends should be strongly amplified if a trend is seen, e.g. across many people or types of interaction as this usually indicates an overall loss of community trust in the person. Exceptions to this last, and perhaps a way for members to indicate that changes in friendship should not be incorporated in trust measures, should be studied.
4. Hosting and surfing participation (this is implicit in the friendships etc that are made, but more explicit measures can also be extracted - for instance numbers of surfers vs numbers of ones that left good references)
5. References. We can easily map the current reference system onto numerical values: Extremely positive = +1, Positive =0.5, Neutral=do not consider, Negative = -0.5, Extremely Negative =-1.0.
References should also be particularly searched for keyword/tag search and for supporting measures of profile authenticity (for instance if a person says they are outgoing, and references also mention this, it should be considered in the authenticity measure.. but perhaps that is for later).
6. Shared travel (especially if connected through CS as opposed to already knowing each other)
7. In person meetings that are not surfing, including
- Meeting for coffee or a drink
- Meetups, parties, planning meetings;
- Collectives
(these could all be weighted for duration and/or investment, i.e. an hour-long coffee meetup should be weighted far less strongly than an interaction at a collective)
8. Chat. By far the most tricky set of measures, but crucial because it is often a first point of personal contact for Csers, because many CSers who travel extensively use the system, and because while not in person, it gives a group 'sense' of a person. It is also a place people come for assistance in using the system, asking for advice, or otherwise asking questions from a 'live' person. Perhaps most importantly, regulars on the chatroom form their own trust communities in a very direct and clear way, and can serve as a backbone for extensions of the network and the building of additional trust sub-networks.
9. Measures on profiles:
- 'Fullness'(are all sections filled out? - automate this) or
- Usefulness in conveying a sense of the person (have viewers of the profile do a quick 1-5 ranking of 'how useful is this profile in giving you an impression of the person)
- Authenticity: only for those who have met the person, have them rank how much the impression given by their profile matched the impression received in person.(Could be a -1 (inaccurate representation) to +1 (accurate representation) scale so that inauthentic people can be easily determined and winnowed)
10. Donations, especially volunteering time, skills, etc. Although this is already implicit to some extent in the fact that interactions with other volunteers generate friendships, references, and vouching, this needs to be separate to reflect other kinds of input and interaction. Weightings on the various kinds of donations should reflect the degree of trustworthiness involved in them – that is, unknown people buying their way in should not result in overly large increases in 'trustworthiness', while people willing to expend large amounts of time and energy on the project indicates a type of implicit trustworthiness.
11. Additional measures of evaluation not yet implemented. These could capture information not always represented in the system, for example :
-allowing people to be evaluated in a 'looser' way (and not as strongly weighted) from meetups, chat, emails, etc. That way this information is not lost but rather utilized, and could give an overall view of how, for instance, a group of people in chat or at a meetup perceived a member in a more group-oriented setting. This might be especially useful in deriving some kind of indicator of cultural sensitivity (if all the Brits say the person is fine at their meetup, but all the Indians say no they were not - this could serve as a flag re the lack of cultural sensitivity.
Aims and further ideas
Aims of this portion of the project include: increasing the trust people have in the system, reflecting true degrees of trust people have in others, and propagating that trust throughout network connections. Much of this infrastructure is in place (e.g. the 'degrees of separation' function), and additional software infrastructure can be used in a multitude of ways that would further the integrity of the couchsurfing feedback system.
Again, the idea is to present this in a way that makes clear and explicit sense to members, increasing their own trust in the system. It could also be extended in at least two important ways:
- An active winnowing process of members that don't meet basic 'trust' criteria.
- Eventual anchoring of a member-based decision-making process that honors the massive input and vision of the founders, major volunteers, etc , compared to new or less trusted people.
