November 16, 2012 – A behind-the-scenes provider of advanced navigation technology for the mobile and pay TV markets is hoping to supplant semantically restricted modes of speech-activated search with a much more conversational approach to helping consumers find what they want.
Veveo, a Boston-based supplier that counts Verizon, Comcast, Cablevision, Rogers and DirecTV among licensees of its pay TV search technology, says it has developed a “conversational interface” to its discovery processes that will allow licensees to put commonly used speech activation software to much more effective use on their platforms than has been the case so far. “Our conversational interfaces are a logical evolution of our deep semantic technologies currently being used by Tier-1 operators for natural, intuitive and intelligent search, personalization and discovery features for multiscreen television and video services,” says Veveo founder and CEO Murali Aravamudan.
Slated for demonstration at the 2013 Consumer Electronics Show, the new Veveo SmartRelevance Conversational Platform goes beyond the simple commands common to voice-activated search systems that use pre-programmed categorical references to link speech with content. “This is a speech interface to our semantic search fabric that allows us to simulate natural language dialogs with almost human intelligence,” says Sam Vasisht, chief marketing officer at Veveo.
“Saying things like ‘channel up’ or ‘channel down’ or ‘show me CNN’ is too limiting when it comes to how people are accustomed to describing video content,” Vasisht says. “You should be able to tell the difference between someone saying, ‘I’d like to watch this movie’ or “I’d like to watch a movie like this,’ where the connotations of the word ‘like’ are completely different.”
For example, he says, “What if you could talk to the TV and ask, ‘What’s the latest Tom Cruise movie available on VOD? What year was it made? Is it an action movie or a comedy?’ You’re not saying Tom Cruise all the time. You’re talking naturally using pronouns, not just nouns. This is what we’re bringing to the market.”
In a video demo the conversational flexibility Vasisht describes is illustrated with the give and take between the Veveo platform and a user querying an iPhone for information about what’s available through traditional and OTT video outlets. For example, after a series of questions and answers pertaining to NFL games and who is playing whom, the questioner suddenly switches topics and says, “Show me some action thrillers,” to which the system responds with a set of thumbnails of movies in that genre.
The system then responds with thumbnail options to a flow of inexact questions which normally would draw a blank from a voice search system: “What about Bond?” “Show me some old ones.” “Ones with Sean Connery?” “Show me the ones on Netflix.” At this point in the demo the user suddenly switches topics again, asking, “Are there any games tonight?” and drawing a response showing various types of sports events available. The user asks, “Any Sox games on tonight?” and elicits: “Did you mean Chicago White Sox or Boston Red Sox?”
After obtaining further information about whether the Boston Red Sox are playing, the user suddenly commands: “Play Hunt for Red October on Netflix for me,” at which point the movie begins playing. The key to understanding what is going on with the Veveo conversational interface isn’t the type of voice recognition technology in use but rather the underlying approach to search, whether by voice or text, that Veveo has been delivering and building on for the past five years.
In essence, Vasisht explains, the responsiveness demonstrated with the iPhone app rests on the fact that Veveo, holder of 32 patents with another 34 applied for, employs a methodology more akin to new types of Web-based semantic search such as Google recently introduced than the metadata tag searches common to pay TV applications. “We’re not wedded to any one type of voice recognition technology,” he notes. “We’ve built the interfaces to work with commercially available voice recognition systems like Nuance, AT&T’s WATSON, Google and others. It’s up to our customer to bring the voice recognition. We provide the stack that starts with the processing of speech as it’s converted to text.”
The Veveo search method is based on its patented version of knowledge graphs, which allow search to go beyond simply identifying things by name as they are randomly scattered across the Internet or in TV metadata files. Instead, the knowledge graph is a matrix of nodes representing specific objects, each with its myriad connections to other nodes which have been built from existing metadata and other information.
Such information may have been linked with other pieces of information previously, but it has been done in duplicative ways where there’s no intelligent integration of one set of links with another, even though they may all have one or more objects in common. “The reality is metadata exits in many different places with different levels of accuracy and completeness,” Vasisht says. “What we do is crawl the Web to find authoritative sources of metadata and automatically capture, combine and enrich that information for use in the knowledge graph.”
In other words, the knowledge graph reprocesses all the disparate pieces of data associated with a given object by making the object a node, setting degrees of proximity to ensure the connections likely to be most relevant in a given search are used first. In Veveo’s case, the technology employs more than 100 million “SmartTags” for dynamic reference to named entities in conjunction with mapping and cross-mapping of white-listed authoritative sources on the Internet and within enterprise systems.
In a blog describing Google’s newly deployed knowledge graph solution in May, Google senior vice president of engineering Amit Singhal said the move “is a critical first step towards building the next generation of search, which taps into the collective intelligence of the web and understands the world a bit more like people do.” This is an algorithmically maintained matrix “that understands real-world entities and their relationships to one another: things, not strings,” Singhal wrote.
“For more than four decades, search has essentially been about matching keywords to queries,” he noted. In contrast, “Google’s Knowledge Graph isn’t just rooted in public sources such as Freebase, Wikipedia and the CIA World Factbook. It’s also augmented at a much larger scale, because we’re focused on comprehensive breadth and depth. It currently contains more than 500 million objects, as well as more than 3.5 billion facts about and relationships between these different objects. And it’s tuned based on what people search for, and what we find out on the web.”
Google, like Veveo, has brought speech recognition to the new search model, allowing users to tap into the underlying matrix using conversational language, although, judging by each company’s demos, not to the degree of abstraction apparent in the Veveo app. “We’ve begun to gradually roll out this view of the Knowledge Graph to U.S. English users,” Singhal said. “It’s also going to be available on smartphones and tablets.”
In bringing to bear its own knowledge graph technology to serve various verticals’ needs over the past several years Veveo has leveraged other innovations to augment customization and improve user experience, Vasisht notes. For example, he says, system algorithms map known content catalogs and vertical data spaces to knowledge graph elements to facilitate the distribution of content assets from catalogs in response to queries.
Perhaps most important, Veveo has developed a personalized contextual learning engine that maps individual behavioral patterns and interests to the knowledge graph to enable “hyper-personalization” of query responses. This “personal graph” creates concise individual signatures with spatial-temporal modeling and device awareness based on statistical learning, Vasisht explains. “If someone searches on Tom Cruise, we’re able to predict with accuracy what that person will be interested in relative to Tom Cruise,” he says.
The video vertical has been an important early adoption arena for Veveo’s technology, sometimes in conjunction with enabling search on conventional TV navigation grids and sometimes with next-generation UIs. For example, in March Verizon, which first deployed Veveo technology in 2007, expanded the licensing agreement in connection with settlement of a patent infringement suit brought by Veveo.
Verizon will be able to leverage Veveo’s technology for content discovery across Verizon’s FiOS TV channels, video-on-demand, and DVR content for viewing on TVs, tablets, mobile phones and other connected devices, says Eric Bruno, vice president of product management at Verizon. “Veveo is a leader in providing consumers with new ways to discover video content they want to see,” Bruno says.
As part of the agreement, Verizon entered into a non-exclusive license of certain Veveo search patents for both TV and mobile devices, he notes. “Video search and personalized recommendations are a developing component of the FiOS TV customer experience,” he says. “We are pleased to license Veveo’s technology and begin exploring additional ways to work together.”
One of the benefits touted by Veveo for its platform is the speed with which the system identifies the item the user is inputting into the text field. “Predictive search was a new concept when we introduced it in 2007,” Vasisht says, acknowledging that now many entities offer this capability, though with a different search methodology that limits the range of options that might be predicted.
Predictive search is especially useful in the TV navigation space, where typing letters is difficult. “When people are searching on TV, it’s very tedious to type in every letter on a virtual keyboard,” he notes. “Most searches on our platform are completed within three key strokes and virtually all are completed in five.”
Some Veveo customers are relying on traditional remote controls to perform the text-based search. In these instances search is activated by the user on a designated key, and then the number keys are recognized by the system as representing groups of letters, as on a telephone keypad. The system quickly discerns which letters are intended in a few keystrokes to produce the search results.
Along with video, mobile apps have been a key target for the Veveo platform, which now runs on over 100 million connected devices worldwide, according to company officials. In July the company introduced an app search app designed to make it easier for users to get to what they need when they need it.
The new vtap QuickSearch, as demonstrated on the Veveo website, searches across device content, including contacts, calendar, music, text messages, device settings and other categories, to seamlessly merge with online results from various Android app stores, Wikipedia, Wiktionary, movies, local business listings and places of interest. The app then prioritizes the results based on the user’s learned preferences, including preferences tied to time of day and location.
The Veveo solutions are offered on a hybrid-cloud platform that can be deployed in operators’ or service providers’ networks, private clouds or hosted by Veveo, Vasisht says. Getting the platform up and running within private operator firewalls is simple enough, because “all we’re doing is taking the software we’ve built and running it on their servers.”
“Where the work comes in is building the functionality they specifically want,” he says. “No one uses all our functionality. Each customer chooses the features and how they want to implement them and what the UI will be that goes on top. That’s where the work comes in with the design and integration with their UIs and other navigational processes.”
But even this aspect of implementation is relatively straight forward, he adds. “Because we run everything in the cloud using standard IP technology, integration is very quick and easy for our core technology,” he explains. “We don’t run any software on the set-top, so there’s no need for installation there. The cost and time required and the hardware dependency are extremely limited.”
Still, the move to advanced search in the TV domain is a slow process. While Veveo has gained traction with many Tier 1 service providers, the truth is the technology is racing ahead of where most operators are in the deployment process, Vasisht says. “Nothing happens quickly in the world of operators,” he notes. “Commercially deployed aspects of our technology typically lag where we are in the development cycle by a few quarters or even a couple of years.”