The greatest recent change in the ability to search within social media sites has undoubtedly been Facebook’s introduction in January 2013 of Graph Search, currently limited to selected users of the Facebook US English interface.
To mark Graph Search’s 6 month anniversary, let’s assess in depth what Graph Search does and note a few of the changes Facebook has made since the initial release.
What Do They Mean By Graph?
In social media the relationships between people and between people and entities, such as places, photos, posts or interests, can be represented by a graph. Arrows, called edges, represent symmetric and asymmetric relationships between people. They also show ties between people and things. The shapes which represent people and entities are called nodes. Each node can have its own attributes, such as a person’s home town and gender or a photo’s date taken.
Image source: Facebook
Big Data Equals Big Search Challenges
In trying to make its site searchable, Facebook faces a massive challenge. With over 1.11 billion people, 240 billion photos and 100 billion connections, Facebook deserves a big data label. Moreover these numbers increase daily. 600 million users log in every day and collectively perform about a billion searches (Google apparently processes about 3.3 billion searches a day).
Source: Facebook Graph Search launch press conference
As the quantity of Facebook’s data has increased, Facebook actually reduced a user’s ability to search the data. In the early days users could discover other users based on a wide variety of profile data elements.
Once upon a time: advanced people search in Facebook. Source: Boolean black belt
Facebook user search circa 2012. Source: Antezeta
By 2012 Facebook had vastly limited profile search options. Complexity introduced by greater privacy concerns and the immense growth in Facebook’s user base most likely led to this reduction. One could be forgiven for thinking Facebook was ambivalent about their search function. Behind the scenes search was actually the focus of a major product development effort, as we now know. The result, Facebook’s Graph Search, is truly an amazing accomplishment, even if there is still a lot of work left to do. Let’s see why.
Graph Search At First Glance
At its launch Facebook presented Graph Search as a way to create and label a custom view of Facebook by simply “editing” the page “title” to reflect the view they wanted to see. In keeping with that metaphor, the Graph Search function appeared as white text within the blue “title” bar which already extended across the entire top of the Facebook site. One design goal was to encourage people to enter specific, and often long, queries. Facebook probably wanted to stop many people, including me, from attempting to post a status update in the search box as well. The three notification icons which appeared to the left of the search box were also moved to the right side of the title bar, making way for a longer search string. For the same reason the Facebook name was reduced down to an f icon.
Apparently not everyone grasped that they could type over the Search for people, places and things text to initiate a search. In May Facebook brought back a traditional search box which underwent further refinement by June as Facebook added a magnifying glass and moved the stylized f out of the search box
Try This Search!
To help users understand the range of queries Graph Search is able perform, Facebook provides a drop-down suggestion list as soon as the cursor focus shifts to the search box. The initial list included
- My friends
- Photos of my friends
- Restaurants nearby
- Games my friends play
- Music my friends like
- Photos I have liked
By June two more suggestions appeared
Explore Facebook Graph Search With Even More Examples
Facebook appears to be acutely aware that most users will not initially appreciate the full breadth of what can be discovered using Graph Search. Facebook provides even more examples in their introductory guide:
- People who like cycling
- People who like cycling and are from my hometown
- People who like cycling and live in Seattle, Washington
- Photos before 1990
- Photos of my friends in New York
- Cities my family visited
- Restaurants in London my friends have been to
Although Graph search is currently limited to US English, the Graph Search introduction page is already available in other languages, such as Italian.
In February Facebook posted even more examples. I’m not really sure why we’d want to find friends who are engaged (I would have assumed that people would be looking for friends who are single…) but certainly some of the other suggestions are fun or practical or both.
- Hotels near Museum of Fine Arts, Boston
- Ski resorts my friends have been to
- Places I have visited
- Places that people named “waldo” have been to
Discovering New Things
- Videos of TV shows liked by my friends
- Movies liked by people who like my favorite movies
- Favorite hobbies of C.E.O.s
- Apps my friends like
Professional Use Cases Too
In addition to Facebook’s official suggestions, there are myriad use cases applicable to the business world, particularly in the areas of sales, recruiting and supplier research. Just consider:
- My friends who work at HP
- Friends of my friends who work at HP
- People who are not my friends and work at HP
Strengths And Weaknesses Emerge
Many of Graph Search’s current strengths and weaknesses emerge through the above queries. Graph Search correctly recognizes “HP” as a synonym for “Hewlett Packard”. It also understands
- My friends who work at HP
- My friends who have jobs at HP
as equivalents. Yet many Facebook profiles I’m familiar with list a person’s current employer as their previous employer – apparently a problem of bad data. I imagine this is due to information entered before a certain date, although I have not investigated this in depth. Certainly I’ve also seen many cases of users incorrectly entering data, such as inverting their first and last names. Currently there doesn’t appear to be any way to perform a more complex “OR” query to search on people who currently work for a company or previously worked for a company. Even if such as query were possible, the answer would only be partially correct.
In another example of what appear to be erroneous results, I searched for My friends who work at NTT DATA Italia. The query does appear in the list of suggestions and I know of at least two profiles I can view which should match the query, yet I get 0 results – I suspect as they both work for multiple companies, at least according to their profiles.
Unique Results Page URLs To Share Searches
Each Graph Search query results page has a unique URL, making it easy for users who have Graph Search to share a query results page should they so desire. There are a few caveats. Users who don’t have Graph Search enabled see a graph search promotion page. The link to search for nearby restaurants will bring up restaurants in Milan Italy as Facebook has converted nearby into a geographic place which will vary based on a user’s location.
Dealing With Duplicate Matches: Disambiguation
To date most people have probably used Facebook’s search function simply to find people by name. While I’ve had access to Graph Search since the end of January, it took a few months before my searches evolved beyond first name last name. Facebook clearly wants us to understand that it has a whole world of entities, tied to us, to our friends and to friends of friends, which might interest us and keep us further engaged on Facebook. Yet interpreting a user’s intent and presenting them with choices isn’t an easy task and despite great strides, Facebook knows it still has some work to do. Consider a search for My Groups. Two exact matches appear as suggestions.
The first option displays a Graph Search results page of My Groups; the last option displays the classic Facebook group listing, a difference which probably isn’t immediately clear to the uninitiated.
In other cases Facebook may prompt the user to help better clarify the search intent, something I don’t remember Google ever doing.
Mobile Support, Status Updates And App Activity Are Missing
Facebook said their initial focus in launching Graph Search is on people, photos, places, and interests. News feed posts and comments are planned, but for now are conspicuously absent, although Facebook’s introduction of hashtags in June can be seen as a first step in the support of post search. Similarly Graph Search is unaware of App activity. Graph Search will eventually be made available in other languages too, but given the linguistic complexity involved, it probably will happen over years rather than months.
Despite the importance of mobile devices to many Facebook users, Graph Search is not yet available on mobile devices, not even in the mobile web browser interface. Mobile will present many design issues including difficulty in entering a long query, the limited screen width to display suggestions and the poor quality of mobile connections which will delay Facebook’s displaying of search suggestions.
Examining Graph Search Results
Facebook Graph Search results snippets. Source: Antezeta
When Graph Search returns multiple search results Facebook provides a contextual snippet, based on the query, next to an image for each result. The snippet aims to help the user understand how the result is pertinent to the search. Each search result also contains a magnifying glass button which seeks to encourage users to explore items related to the search result, such as photos or the person or people who visited a place. The choices, which only appear when the mouse hovers over the magnifying glass as in the first example above, depend on the type of search result. This is a really nice touch which I suspect people will only discover over time.
Results are ranked by a number of factors which may include the relationship between people (a relative is closer than a friend, a friend is more important than a friend of a friend), the number of mutual friends and other signals of affinity such as city or engagement (likes, comments, ratings). It is reasonable to expect that Facebook will be evolving result ranking factors over time.
The results Facebook returns not only aim to match what the user expressed, but what was intended. A query for single women who live near me or single men who live near me, as the case may be, would ideally try to understand that the user is intent on dating and is probably looking for people of the same age and sexual orientation. It seems though that Facebook still has a lot of work to do in correctly interpreting a user’s intent. Special rules apply to teenagers. They may show up in just their friends queries, or friends of friends of the same 13-17 year age range if Facebook thinks the search could help to identify a young person by age or by their location,
When a user searches for people at companies or educational institutions, Facebook will try to give greater weight to users who have verified email addresses from the company or school. Facebook wants to distinguish between those who really went to Harvard from those who just say they did. A current search for People who live in Italy and work at Facebook yields more than 1000 results, a rather implausible scenario.
Rather surprisingly, there isn’t yet a way to sort search results.
Search Results Refinement Options
Contextual refinement drop-down list filters. Source: Antezeta
Contextual refinement drop-down list filters appear to the right of the search results. Note that these are not comprehensive, even if a user clicks on “See more”.
Extend this search. Source: Antezeta
Also part of the search refinement area are options to Extend this search, which provides related search suggestions to dive into information about the current result set and, after a block of advertising, to Discover Something New. The advertising block was introduced in April 2013.
Desktop searches using nearby, such as restaurants nearby, will depend on IP geo-localization which is notoriously unreliable at any level of precision beyond country. There isn’t much Facebook can do about this other than asking a user to specify their current location. Mobile devices using GPS location services should not have this problem once Graph Search appears on mobile devices.
No Keywords Here, Move Along, This Isn’t Web Search, Our Data Is Structured
In designing their new search system, Facebook has a massive advantage over companies like Google and Bing which offer traditional web search. Facebook’s data is relatively clean and structured, especially since Facebook began to use Wikipedia as a data authority. Facebook knows that every person and entity has specific attributes which have specific values. For example, the types of relationships a person may be in, e.g. Single, Married, Widowed, are well defined by a finite list.
Facebook was thus able to design a powerful search function which uses a precise query to directly deliver what should be relevant answers opposed to the traditional keyword based web search which returned links to things we still need to explore. Facebook’s answers are the people and other entities to which we are often connected to, directly or indirectly, in Facebook.
Google might argue that web search provides direct answers too. Through Google’s Knowledge Graph, Instant Answers and Google Now, Google too has been moving away from the old 10 blue links of the past.
Facebook Graph Search Is Still A Little Rough Around The Edges
Facebook graph search is undoubtedly a major accomplishment. There are however areas where the many challenges facing Facebook become even clearer. User query interpretation is still problematic. As any professional working in the search industry knows, users never cease to surprise with the multitude of ways they find to express a concept. To work well, Facebook’s Graph Search must have robust matching for synonyms and related concepts – whether expressed as a single word or as a phrase – in each of the more than 70 languages it will support, not a small challenge
Synonym And Phrase Equivalents Recognition
In the business related search examples noted previously, Facebook demonstrated some success in interpreting synonyms and phrase equivalents. They do this on many other areas too. Typing Films produces the suggestion Movies and correctly expands it to Movies my friends like should the user type Films my friends like. That isn’t always the case. Want to discover a new TV series based on what your friends like? At the time of this writing, Facebook is aware of TV shows, but not TV series. If you type too quickly or paste TV series into the search box, you’ll miss the suggestions for queries Facebook can handle which appear after typing TV s… To make Graph Search work, Facebook is in essence also constructing a database of alternative ways to express a concept, a task which needs to be done for each of the languages Graph Search will support.
Inexact Name Matching
In people search, I found that typing a first, middle and last name would not match on just the first and last name if the user had not specified their middle name in Facebook, although this problem seems to be limited to non-anglo saxon names. Since I started using Graph Search, I do find myself using Google more often to find people on Facebook: not a good sign.
Query refinement, the modification of an existing query, is also problematic. I often find myself searching on the name of someone I’ve met at an event, selecting one of the names only to realize I’ve chosen the wrong person. I go back to the search query box where Facebook displays the name of the person I selected. The other people to choose from are gone. Facebook’s search history has apparently kicked in, thinking I really just want to see the guy or gal I just selected. The natural reaction would be to hit the backspace key, thinking that might remove the last letter from the current query, effectively telling Facebook to show all matches for that almost complete name. Unfortunately Facebook uses the backspace key to clear away the entire search string. A power user eventually learns that they need to move the cursor to the penultimate letter of the search name then delete the last letter. I cannot imagine that I’m the only one who has encountered problems with this people search behavior, yet 6 months on the behavior has not changed. To be fair, I maybe need to change my behavior to chose the suggestion “Find all people named [so and so]” (back to the 10 blue links scenario) rather than selecting a person at a time.
Engaged Users Will See Richer Results
Facebook Graph Search results which consider data from a user’s profile, i.e. my friends or my likes, will be depend significantly on the degree to which a user is engaged on Facebook. A user approaching Facebook’s 5,000 friend limit will often see richer Graph Search results compared to a rather shy user who has only friended a few people and liked a few things.
What Should Businesses And Organizations Be Doing?
Facebook’s advice to businesses and organizations is simple: work on your Facebook page. Start by ensuring basic page information is optimized for Graph Search:
- The page name
- The page’s category
- The friendly (vanity) URL
- The “About” text
Address information, if appropriate for an organization or business, may be used by nearby and other location based queries.
Fan data, including the number of fans and the level of fan interaction, can be a Graph Search ranking factor. As in real life, not all fans are created equal.
Graph Search does now support hashtags, something to keep in mind when posting updates.
Different views of what privacy means and the inevitable tension between individual and commercial goals have made privacy one of Facebook’s thorniest issues. Improved search within Facebook naturally enables surfacing of information which was already visible in theory, just difficult to find. In releasing Graph Search, Facebook has tried to address privacy concerns head-on. Prior to the Graph Search release, Facebook launched new privacy audit tools. In its Graph Search announcement, Facebook stated:
We’ve built Graph Search from the start with privacy in mind, and it respects the privacy and audience of each piece of content on Facebook. It makes finding new things much easier, but you can only see what you could already view elsewhere on Facebook.
Source: Facebook Graph Search launch press conference
Facebook later followed up with a post on privacy, a dedicated help center article and a page in the about Graph Search presentation. As some have noted, by default Facebook maintains a record of an individual’s search history. In theory this remains private, but accidents will happen. You can delete Facebook’s Graph Search history, but this may impact Facebook’s ability to tailor search predictions and results to your intent, as expressed by prior searches.
To be fair, Google also tracks a user’s search history in order to better tailor search results, but Google is much clearer as to what it is doing and how a user can disable search history logging. Facebook will help a user delete one search from the search activity log, but I could not find an official help article on deleting (Facebook calls it clearing) the entire search activity log. Nor does there appear to be a way to turn off search history logging, which in my view opens Facebook open to criticism which could easily be avoided.
The Bigger Picture: Where Graph Search Fits Into The Facebook Ecosystem
During the Graph Search launch press conference, Facebook said that there are three pillars of the Facebook query ecosystem:
- News feed answers the question what’s happening around me.
- A person’s timeline answers the question: who is this person.
- Graph search is for everything else, particularly helping people stay connected and discover new connections.
Behind The Scenes: Understanding User Queries
One of the impressive feats of Graph Search is its ability to support a robust natural language query syntax. A user should be able to tell Facebook in detail what they’re looking for, without worrying about learning a special search syntax. In a Reddit Ask me anything session, Graph Search co-designer (and ex-Googler) Lars Rasmussen said Graph Search’s query language works on three levels. The first is the natural language that users type into the search box. Facebook then interprets this into a semantic language that describes its understanding of the user’s query. Finally the technically inclined might be pleased to know that Facebook uses an s-expression syntax to retrieve potential results from an inverted index.
Implications For Google And Others
Facebook is quick to stress that Graph Search isn’t web search, Graph Search is limited to data within Facebook. Yet as noted above, Facebook makes extensive use of Wikipedia to populate pages within Facebook, vastly increasing what Graph Search can find inside Facebook. Graph Search also falls back on Bing’s web search when an answer isn’t found in Facebook’s own data (in Italy this doesn’t currently work, I assume as Bing Italy uses a country specific URL, it.bing.com, instead of the usual www.bing.com). Thus users should be able get Graph Search to answer many of the questions they normally pose to Google. Less clear is what impact Graph Search will have on specialized sites such as LinkedIn. It isn’t a far-fetched idea to imagine Facebook complementing LinkedIn as a recruiting tool.
While it is unlikely Facebook is going to become a destination search engine, I would expect many use cases where people will decide to take advantage of the Facebook search box already in front of them rather than going to Google. Despite Facebook’s admirable efforts, user adoption of new search tactics to fully take advantage of Graph Search’s rich features will probably take time.