Keith Griffin describes how emerging social and graph data technology can remove barriers to more effective collaboration.Interview conducted by Alan Morrison and Bo Parker
Keith Griffin is lead architect for Cisco Systems’ enterprise collaboration platform business unit. His background includes 15 years of experience as an architect and development engineer at Cisco and Nortel Networks in the areas of unified communications, IP telephony, call center software, and web-based collaboration.
PwC: How did you initially get involved in collaboration software?
KG: I was one of the founding members of the unified communications collaboration research and development site that we have in Galway, Ireland. That site focuses primarily on the unified communications products, such as soft clients for voice video, instant messaging presence, federated learning systems, and integration with other unified communications systems.
PwC: Cisco is known as an IP router and switch company. Why social software?
KG: When you look at our architecture side, quite a lot of assets are already in there, either through acquisition in the portfolio or from existing products— such as presence and location, the client services framework for communications, the ability to record, but based on top of the networking layer that Cisco is known for.
Then we also identified that things like policy, for example, could be used. If you consider some of the well-known social networking applications, they’re really about connecting with friends. In the workplace, you also choose how much information you want to share or not, but the social networking applications are not really about friends, friendships, and the enterprise. They’re about getting something done.
So if you want to get something done, you have to have policy rules and security. Somebody might want to share an idea with a board or a council, but that idea might not be appropriate to share with the entire organization until it is a well-formed idea. You can apply policy rules to say what information, what communities, and what people can share.
Again, we had assets in this area from an acquisition, but it’s an XACML-based policy engine—an XML-based standard for sharing policy. So we were able to use that again in the platform to enable policy.
Then, on top of that, we knew we would need to develop some new capabilities, such as tagging, social graphing, semantic processing, and search. We don’t necessarily intend to be in the search business. That’s not the goal. The goal is to use a search engine and couple that with some of the capabilities involved, like the graph and the semantic processing.
If you want to get something done, you have to have policy rules and security. Somebody might want to share an idea with a board or a council, but that idea might not be appropriate to share with the entire organization until it is a well-formed idea. You can apply policy rules to say what information, what communities, and what people can share.
PwC: Search would seem to be a necessary component.
KG: Yes, it absolutely is. But we also want to look at where we’re coming from. It’s very important that we focus on the areas where we make a difference immediately. Most organizations— not everyone, but most organizations— will have some form of search capability. We decided to make search a pluggable component to the system, so that it can either be a default offering that we create ourselves, or allow search from another vendor to be plugged into the system. We thought that was important.
PwC: How do you treat e-mail in systems like this? Our experience with social activity platforms has been that they compete with e-mail. If just a few people stick with or revert to e-mail, then a conversion to something different just doesn’t happen.
KG: These are not things that make e-mail go away, but you could see a marked decline in the need to use e-mail so much because you have so many other ways of interacting.
PwC: Should a wiki be built into tools like these at some level?
KG: We created a social content model where you post information in a flexible form. You don’t necessarily say, “I’m going to post that to a wiki, or to a blog.” You post to a library and then the library can be shared using the policy model. This method frees organizations from committing to a blog and then struggling with the fact that that information is stuck as a blog post. They need to cut and paste the thing if they want to move it to a wiki, for example.
You can start with a rough note that’s just in your own journal. You can share it in a group of close associates to elaborate. You can hit the share button and then decide to publish it in one place or another, or change your mind later. It brings up all the different communities. If I go farther down the list, there are all the individuals who are part of my social network, who are part of open and restricted communities.
PwC: What’s the governance model for this kind of sharing?
KG: When you publish something to your library, access to that information is governed both by what you decide and what the policy is for that information. The information can show up in multiple places at once even though there’s only one copy of it, right? The copy itself doesn’t have a location, per se. You can put it into the community of your systems engineers or your product sales specialists or whatever, without making a copy.
PwC: What about finding things serendipitously? Can a system like this deliver information you didn’t ask for and weren’t searching for, but the system says, “Based on some pattern, you really should be paying attention to this person, or you might want to follow this topic.”
KG: We have the concept of the activity list and the watch list. A person on my watch list can suggest things I should look at, or the list will have something I’m actively working on.
PwC: What’s the role of triple store technology in systems like this?
KG: Everything that’s going on must be RDF-ized and put into the system so that you build a graph. [In RDF-ization, or conversion to Resource Description Framework form, data is transformed into subject-verb-object triples for more scalable linking at the data element level with the help of domain and element relationship descriptions— richer metadata, in other words. See the article, “Making Semantic Web connections,” on page 20 of Technology Forecast, Spring 2009, for more information.]
PwC: Data in graph form should help with relevance, shouldn’t it? When something new happens in the world, you want to be able to navigate to it very quickly. The system should alert you to new relevant developments as knowledge evolves. How do you treat social and knowledge graphs? Are they treated separately?
KG: No. They’re combined. We haven’t created features in this area, but the design of this graph is broader than a social graph. In our original design philosophy around people, communities, and information, both information and people have some knowledge. Communities are where you may contain some of that knowledge in a persistent way. Our graph is not inherently social. It’s just an arbitrary scope that we placed on it initially.
PwC: If social tools are going to be successful, useful, and not just add to the workload but actually subtract from the workload, you need passive and active expansion and trimming processes. How do you use the social platform to enhance the processes around the workflow?
KG: One of the reasons for choosing a Web 2.0 and widgetized type of deployment is the concept of contextual awareness. The application should be aware of what I’m doing now.
One of the things that’s come out of Semantic Web in the last few years is Linked Data. Linked Data is really an excellent way to provide a foundation to make contextual awareness feasible. Building a system from the ground up allows technology choices. You aren’t tied to, let’s say, a relational model. You don’t need to create a graph using tables in a traditional database. It actually is a graph. That makes a big difference. That gives you the flexibility to do the kinds of things you’re suggesting.
From a user experience perspective, you can take a conservative approach. You can avoid getting into advanced visualizations and delving into contextual awareness. The first step is to offer an Enterprise 2.0 platform. Once you bring value and get people on that type of platform, then you can introduce these more intelligent features and start making life easier.
PwC: How does the graph model affect enterprise search? Relational databases, by contrast with graph data, are about individual records and summarizing a bunch of records, not navigating through a data space.
KG: We call the concept that you’re describing relational navigation. If I search for the term “Semantic Web,” a system with relational navigation will give me that three-dimensional view of people, communities, and information, so there’s the people, there’s the communities associated with it, and there’s the information in the center. You can bring the social aspect to the search as well as the unified communications.
The difference the additional social graph information brings is that the activities users are mentioning in the stream are now the activities of the community. It’s not just random activities and people who happen to be associated with the community. These are people working on this topic right now. It’s an active and current thing.
PwC: How does the social graph lend itself to applications development?
KG: You can connect to your LDAP [Lightweight Directory Access Protocol] directory and get started, or treat these systems as platforms on which you can build. Every single early customer that we’ve had has done work on a socialenabled business process or tool for the platform.
The industry can offer open interfaces through developer networks. With these interfaces, a university such as Duke can create certain kinds of libraries to describe a classroom, a study group, a teaching assistant, or a professor. [See the interview with Tony O’Driscoll of Duke University on page 18 for more information.] It’s a business process that is social, but it’s built around the business process.
I’d make a distinction between what an API [application programming interface] such as OpenSocial can do, and what you can do by querying the graph directly with SPARQL [SPARQL Protocol and RDF Query Language, part of the W3C Semantic Web stack; see the article, “Spinning a data web,” on page 06 of Technology Forecast, Spring 2009, for more detail on SPARQL]. The system we use internally supports both methods. An API is constrained by what it’s able to do. With the OpenSocial API, you can make only the calls that the API allows you to, but it is designed for systems that have the concept of social. So if I wanted to find out how many of my friends of friends have posted content on a particular subject, I could do that with OpenSocial.
In Andrew McAfee’s book on Enterprise 2.0 [McAfee is a Harvard Business School professor and the author of Enterprise 2.0: New Collaborative Tools for Your Organization’s Toughest Challenges], he talks about the fact that in the social network, your direct friends are at one degree of separation. If all you do is spend time making the relationships you already have stronger and ignoring the ones you don’t yet have that potentially have more information and more use for you, what you’re doing has limited value.
But if you can see activities at more than one degree of separation, a friend of a friend or a friend of a friend of a friend, then you can create very interesting applications. You want to create those applications, but you also want to open the API so that people who want to do this in their own setting in their own environment can do that as well.
PwC: You could have two organizations but because they define policies different ways or they’re using the thing in a different way, they can’t actually combine communities across enterprise boundaries, which is what you really probably want to do in some cases.
KG: That policy control is critical, and it’s one of the reasons that the focus is so much on federated social software among different groups.
We’ve seen some of the leading research groups in the world look at platforms like this, where they might be doing joint research projects. Maybe a major UK university has some funding and is going to bring in a major US university for six months. They want that temporal view, so they have access to this community for six months.
Maybe they want to bring in a university from China or from somewhere else and set up a longerterm federation between both systems where they maintain their independent systems but they share information. I think there’s going to be a lot of opportunity for standardization. You can federate at a protocol level. Some platforms can federate very well with things like SIP [Session Initiation Protocol] and XMPP [Extensible Messaging and Presence Protocol], but there’s social federation, too.
PwC: What have your customers been interested in doing with this sort of platform?
KG: It’s not all to do with social software directly. For example, if somebody wants to change how their intranet works, they look at it from a completely different perspective. That’s one of the things Cisco itself did. Our intranet of applications grew over 10 years and often not with a lot of structure. So Cisco changed the user experience.
PwC: In a sense, it’s a redo of the home page.
KG: Yes, but now it’s widgetized. And this is where governance comes in once again. We have our internal browserbased home page, but then we do all of our work in other applications. They’re displayed close to each other, but the two don’t even talk to each other. For example, we front-ended our engineering document control system with a gadget-type definition as a unit of work or a unit.
PwC: So if you happened to be an HR employee you’d have a different view, one that’s tied to the HR system instead of engineering.
KG: Exactly. This is a major step to move from the Flash intranet to the Web 2.0 enterprise.
PwC: One last question: What’s emerging in this space that hasn’t really been harnessed commercially yet?
KG: Have you seen Sindice [http://sindice.com/]? It’s a Semantic Web index developed at DERI [Digital Enterprise Research Institute]. It ’s coupled with the Sig.ma presentation tool [http://sig.ma/], which builds a social information mashup live from the web. For example, if you search on Stefan Decker [DERI’s director], it should pull in his picture from LinkedIn. It’s not like a directory where you’re going to get whatever is published in the directory. It’s getting whatever is on the web right now, and it can assemble pages on the fly from individual data elements because of the power of the RDF or Linked Data graph. That kind of dynamic publishing capability is at a much more granular level than RSS feeds. So if Stefan Decker published something at a conference last week, it should be in that mashup now.
Sindice is crawling, and it’s RDF-izing. It’s creating triples from every site it can get its hands on, and it’s pulling that in. It’s a demonstration of the plumbinglevel semantic interworkings that are emerging on the web. Of course, this has some obvious enterprise implications. Just think about the ability to pull in individual data elements from LinkedIn profiles (or any external source) and combine them with your company’s own internal profile information and also the social information or stream activity, all of which is referenced via the corporate directory. Once you’ve set up the linkages and domain descriptions, the profiles can be automatically updated, and the system can infer new linkages, too. You’re actually using more information—the richer metadata—to link data silos together.
These are just technology examples. It’s not that these things don’t exist. It’s just that you need to map what’s available and what’s capable, and also consider what the user is capable of dealing with. I’d look at something like Sindice, and I would say that it’s fine. I could probably navigate my way through it, but I shouldn’t need to navigate my way through it. It should happen very naturally.
So we continue to do our user experience studies and research to find the most optimal way to do this.