Frrole : The Official Blog

“Social media isn’t about creating a great brand. It’s about acting as if you already have one”

A few years ago, the biggest concern for digital marketers was to go after numbers. Metrics like no. of followers and percentage reach were the standard yardsticks of success. Not anymore. The world has moved beyond dumb digits and the tougher nut to crack today is to decide what to do with these millions of followers – how to pamper them, how to create strong relationships with them and become the brand they trust and endorse.

The need of today’s marketing is audience engagement, not just to establish brand recall and loyalty, but also from shorter term tactical perspective. Online brand interactions are among the biggest factors affecting people’s purchase decisions today. A report by Chadwick Martin Bailey suggests that 64% people have purchased a product based on social media content, and 50% and 60% people are more likely to buy from and recommend a brand after they follow it on twitter. What this means is that unless brands are actively monitoring and participating in the online discussions about them, they’re ignoring the opportunity to convert highly relevant prospects, sometimes at the point of sale.

Today’s customer is vocal and demanding. He wants to express himself and talk both good and bad about the brands he consumes.  It is both an opportunity and a threat for marketers, depending on the way they harness these social conversations. And it takes more than just posting updates and contests on social media pages to really convert this into an opportunity. Rather, it requires them to create a more immersive and exclusive platform, where their fans can engage, know more about their brand, give feedback, post complaints and talk to influencers directly. Brands from around the world have seen upto 10 times more user engagement, when they integrate social into their offering.

Social data is the best catalyst to encourage user activity, only if one knows how to use it. Digging through the mountains of social data, and extracting precise content out of it can be a daunting task, even for an intelligent analytics software. Even if someone is able to find partial content using #tags, it is still a challenge to use it effectively to achieve desired objectives. Frrole tackles these problems brilliantly with its arsenal of products, that help brands create visually compelling experiences for their users, fueled by highly relevant social content.  

Social marketing is not yet a tool that can replace other forms of marketing, but it is slowly becoming an essential aspect of the media mix that marketers need to build long-lasting relationships with their customers. If used effectively, it allows brands to treat their customers the way they expect to be treated, resulting in stronger bonds and brand loyalty. And needless to say, it automatically takes care of all the metrics, with the numbers representing their real values. 

Nishith Sharma

@nishithsharma

Frrole, a social intelligence startup that helps media and brands mine deep insights and precise, curated content from social conversations, announced its enterprise offering targeted towards the media and entertainment verticals. The Twitter data based offering has three components—data APIs allow clients to access Frrole data streams and use it to build or plug into any application, a Visual Suite allows clients to select and integrate plug and play display applications into their products within minutes and Solutions allow Frrole to work closely with clients for a large project that requires complex data interactions. Frrole is part of the winter batch of the Microsoft Ventures Accelerator in Bangalore, India.

Announcing this major initiative, Frrole CEO and co-founder Amarpreet Kalkat said: “Over the last 1.5 years, we have built a powerful semantic engine that  algorithmically extracts meaningful, relevant and informational social insights from Twitter conversations.  Today’s launch is a major milestone for us and for the $1+billion and ever growing media and entertainment industry as it will offer them unique and innovative ways of maximizing the return on marketing spends. Customers can now use Frrole to extract curated, precise content related to topics, location, products or campaigns, and integrate it within their applications, e.g. to display tweets/ photos shared by users during a product launch. They can also derive insights from social conversations, for example to determine who is the most talked about celebrity in India in real-time. And by using Twitter as a communication channel, our customers will be able to amplify their messaging and reach manifold, for example a TV show can now ask people to vote using a hashtag instead of a traditional SMS.

Congratulating the team on the launch, Ravi Narayan, who drives Microsoft Ventures Accelerator in India, said: “Social integration provides a huge opportunity for brands and media to engage and interact with their users, when they are most vocal.  Globally, companies have been able to achieve 2-10 times marketing ROI by integrating social content and insights into their businesses.”

For example, when GMA News created an online and on-air social experience to cover Philippines 2013 elections, it achieved record breaking 21 million page views, 4 times more than the second place competitor. Similarly, X Factor witnessed 600% jump in twitter activity after its social integration, achieving 11% share of voice and becoming the most social TV show. 

“Frrole has the expertise in identifying, filtering, integrating and monitoring the data and insights that media and brands cannot do without in today’s socially connected world. I am super excited about this new offering, and wish the team all the best,” Narayan added. 

While Frrole’s current focus will be on the media and entertainment industry, this product can be used equally effectively by brands as well. Frrole also offers a very attractive and flexible pricing model, depending on duration, complexity and amount of data needed, making it affordable for enterprises of all sizes. Frrole’s enterprise site can be accessed at http://frrole.com. 

Meanwhile, Frrole’s existing offering for consumers, its Twitter based news discovery site continues to maintain an impressive growth rate. Since the beta launch in April 2012, it has grown to more than half a million users month on month. Frrole mobile apps, launched early this year, have already seen more than 5,000 downloads. Started with just 20 cities, Frrole now covers more than 55 cities in US, UK, Canada, Australia, India and Singapore, and soon will go up to 100 cities worldwide. The news site can be accessed at http://news.frrole.com. 

About Frrole

Frrole is a social intelligence company that helps media and brands mine deep insights and precise, curated content from social conversations.  Using a powerful semantic engine, Frrole algorithmically extracts meaningful, relevant and informational social insights from Twitter conversations, which are further provided to clients using APIs, and are used to build visually rich plug-n-play apps.

For more details, please contact 

debjanim@microsoft.com/ +91 98494 38749

nishith@frrole.com/ +91 993 006 3404

At Frrole, we build great things. But for all this while, the team has been working in isolation, almost entirely living within its own microcosm. While it hasn’t exactly been an impediment to our progress, we have talked often about doing things more ‘visibly’. Especially given the fact that we are getting ready to launch an enterprise offering.

The call from Mukund that way had perfect timing, and he asked if we had interest in being a part of the third batch of Microsoft’s accelerator program. We talked to a few people, some of whom had been part of the earlier batches, and it seemed like an easy decision to make. The program promised great opportunity to meet potential clients, mentors, investors and an extended team, and those were exactly the kind of things we were looking for.

We hear that around 1000 startups applied for the program this time, and after multiple screening rounds, 20 of them were shortlisted for a presentation round. We were the 2^nd team to present, and it went smooth. Finally 10 startups made it through, including your beloved Frrole :)

Another thing happened during all this while. Abhishek Vaid, our coding genius, broke his shoulder while riding his newly bought (old) bike. As we speak, his right arm is plastered but his zest to code has remained unaffected. Amongst us, he was the one most excited to be a part of this program, though he is now trying to use his broken hand as an excuse to shy away from the party that he promised to throw if we get in;-)

As is expected of a good incubator/accelerator program, it is a grueling and fast-paced setup where growth is measured not in months, but in weeks. That is the way we work anyway, so we are more excited than anxious about the commitment needed. Moreover, In the introduction call, we were assured that there will be enough partying, and it is adding to the whole looking-forward to it thing.   

Today, we are sitting here at the Microsoft’s Lavelle Road office, just getting started with the program. In my next post, I will write a little bit on our first-hand experience at the program. And definitely about our enterprise products and solutions, and the verticals we will begin with. Till then, happy frroling! Hope to get surprised with some interesting additions to the consumer site soon.

Nishith Sharma

@nishithsharma

(During late March/early April, I spent 3 weeks in Bay Area/New York/Los Angeles (Frrole is otherwise based in Bangalore). We learnt some lessons, many of which would have been good to learn sooner. Sharing them here, hopefully they would be some help to others starting/undertaking their entrepreneurship journey.)

If you do not have the time to read this whole post, then the summary of this post is: Unless you have a real good reason to be in India/anywhere else, pack up and get to the valley. Or at least NY.

Whatever it takes.

For those with a little more time to indulge, here are a few specific thoughts.

1. The valley has infectious energy. In cafe after cafe, there are scores of people sitting, sipping their mochas and building stuff. Just as you see services companies omnipresent in Bangalore, row after row of buildings have startups lined up one after another in valley. Now a valley native would tell me that I am getting into a bit of hyperbole, but I guess you get my drift. The place is brimming with startup stuff, and given the low cost to startup these days, it just makes so many more people take the plunge.

2.The valley hangs out. My silicon valley startupdigest newsletter would have 30 or 50 events every week, and that is in addition to the tens or hundreds I could see on meetup.com. In Bangalore, we are lucky to have 5 events a month. Now I have never really been an event attending type, and 98% of them are a waste of time anyway, but the other 2% open up brilliant opportunities. 

3. The valley moves. We had more shots at success in 3 weeks in the valley than we probably had in 1+ years over here in Bangalore. One meeting leading to another, introductions, people finding time within hours or days, it was more often the norm than the other way around. In India, I can think of only one investor who perhaps moves at the valley speed. Anybody else, and the threads here do tend to take many months.

4. Money does not grow on trees, even in the valley. In some corner of our mind, we were hoping that people would be sitting out there with checkbooks drawn. That, alas, is not the case. I have been one of the voices myself who berates Indian investors on their inability to look beyond monetization and understand product/team value, but much to my surprise, things out there weren’t that much different. Yes, people did understand ‘consumer’ a lot better, but monetization was still a central theme, especially in these ‘Series A crunch’ times.

5. Know who you are talking to, and talk to the one who knows. Frrole is a consumer media company and a key part of our plan is enterprise as a delivery channel. Whereas many of our discussions valley side hit speed bumps when we talked monetization via enterprise, the speed bumps magically seemed to disappear as soon as we hit New York. For obvious reasons, people there seemed to know what media companies need and how our offering would fit their needs. All old timers talk about ‘rather meet one investor who knows the space than 100 who do not’, I cannot stress that enough myself.

6. The dice rolls differently in the valley. Startups never have great odds, but whatever they are, your odds at succeeding are 100x more in the valley than say, in Bangalore. Especially if you are competing with companies based in US. It might not be terrific to move from .01% probability of succeeding to a 1% probability, but then, it might just be.

Amarpreet Kalkat

@amarpreetkalkat

Hey guys, been long since the last time we met. Such are startups, weeks pass by between two breaths and you never realize how fast time goes. You are the coder, designer, marketer, recruiter, planner and much more, sometimes multiple hats at the same moment. Imagine being in a roller-coaster for weeks, thats how it is. So, I hope our absence for last few weeks will be forgiven :)

On the updates side, we have solidified clusters, grouping all similar and related news items and displaying them within an independent tile. So, now if Mr Kim Jong declares war against South Korea, you won’t be flooded with 20 independent news pieces on the whole page of Frrole. What instead you’ll see is a neat rectangular box, covering one fourth of the screen, and containing everything about this news piece.

Though Abhishek introduced clusters in his last blog post, they were in their initial stage back then and were unable to group each similar piece of news. Over the last one month, he worked day and night, ferreted out whole of internet, dozens of books and improved clustering to the point where we are now able to identify and group even 200 news pieces in a cluster. More about it in the next post by Abhishek.

In the meantime, Amarpreet visited US for 3 weeks and met hundreds of very interesting people from the silicon valley startups, investor fraternity, journalists and a surprise rendezvous with Prof Jeff Jarvis of City University of New York’s graduate school of journalism. More about his adventures in his next blog post.

BTW, we are now looking at opening up our APIs for content based websites for them to be able to use our content in a curated stream form on their websites. We have been getting mails from people asking us whether they can use our data, or if we can help them generate relevant leads for their businesses, and that encouraged us to take this initiative. A few big publishers have already shown interest and hopefully, very soon you’ll see Frrole generated content on our partners’ websites as well.

On a lighter note, the Frrole team is getting stronger on cycling and has moved from 60 to almost 100 kms rides now. Learning from Jack’s mistakes, we’ve made it a rule to work hard but play harder :)

Happy Frroling!

Nishith Sharma

@nishithsharma

Hello Everyone! 

It has been a pretty busy month at Frrole. We’re working towards putting together our technology infrastructure which should make the user experience at Frrole more productive and relevant.

Our major technological challenge was to be able to detect “Duplicate tweets”. That is, tweets which are para-phrased versions of each other. We wanted some mechanism to cluster tweets that talk about same topics or facts in similar context(s). The problem actually is harder than it appears. A temporary solution can be to do a simple keyword based match after tokenizing the tweet’s text and then rank the results. However, doing this becomes difficult since tweets normally possess custom hashtags, user handles and other grammatical idiosyncrasies. The situation is also exacerbated by the fact that any trivial clustering algorithm will take quadratic time to be able to cluster tweets. What it means is that we cannot tag hundreds of thousands of tweets effectively. 

After careful considerations, we came up with a workable solution. We decided to produce a computationally feasible signature of a tweet, which we can compare to obtain an effective measure of similarity. Although finding such a signature is very difficult, we tried to approximate it with the entities which reside in the tweet. For us, a set of entities in a tweet provides a fair basis for its uniqueness. Hence, our first concern was to extract entities from tweets.

Entity extraction, or NER (Named Entity Extraction) as it is formally referred to in academic literature, is a technique to extract real-world entities from a piece of text. This is a fairly standard technique which works as a pre-requisite in many text mining applications. However, performing entity extraction on tweets is challenging because of limited number of words and characters in a tweet. To get a fair idea of the problem, we decided to pursue the research leads of two groups currently working on the problem of entity extraction from Twitter data. One research group is ARK from Carnegie Mellon University and second one is from University of Washington. Both these research groups are trying to find state-of-the-art to this problem. Hence, we decided to take their code-base and build on that. 

We fixed the overall architecture of the project as following:

• Model Implementation: Python 2.7
• Web Framework: Bottle.py with Cherrypy
• Persistence: MongoDB 2.2+
• Core Library : NLTK 2.0+

The reason behind considering these specific technology artifacts was their simplicity in use and deployment. Python being a very versatile language has lots of plugins which makes our task easier and productive. After deciding on the architecture, we started with the main design of our approach. The overall pipeline of the clustering process is as following:

New tweets arrive in the system in batches. We refer to this set as newTweets. All the tweets that are exact duplicate of any tweet in our main_db (Main Database) are deleted.

All the newTweets are pushed into a database called stage_db and a sessionID is generated for this particular session.

All the tweets are processed sequentially, and entitySet for each tweet  is found ( i.e. a set of entities residing in the tweet as reported by NER Engine)

Next, we find the neighborhood of this tweet with help of its entitySet. The neighborhood of a tweet comprises of all the tweets which share at least 2 entities with the original tweet. Any such pair will qualify a tweet to be considered in the neighborhood of another tweet.

After finding the neighborhood of a tweet, we process each tweet in the neighborhood to find Jaccard Coefficient (JC) with the original tweet. Jaccard Coefficient is a standard mathematical tool which quantifies the overlap of objects within two sets. In our case these sets are respective entitySet of two tweets under consideration. In our system Jaccard Coefficient approximates the similarity between two tweets.

Since our ultimate objective is to cluster tweets at any given point in our system, every tweet belongs to at least one cluster. The cluster(s) assignment happens when the tweet enters the system for the first time. If for a tweet a neighborhood cannot be found, it spawns its own new cluster. Moreover, every tweet can also participate in multiple clusters with varying degree of participation. A degree of participation is quantified with a floating point value called participationCoeff, . 

After we have calculated the Jaccard Coefficient(s) of a tweet w.r.t to its neighbors, we need to assign this tweet to appropriate clusters. For doing this, we follow a recursive approach, in which  we find out all possible paths originating from new tweet to all possible clusters via neighbors as intermediary nodes. Obviously, there will be many clusters which could not be reached using this and on the contrary there will be many clusters which are reached by multiple paths. For all such clusters which are reached with multiple paths, we assign the new tweet to these clusters with a ParticipationCoeff as the average of path length, where path length is defined as multiplication of Jaccard Coefficient between the two tweets and participationCoeff of the neighbor in a cluster.

After we have found all the clusters a tweet belongs to, we assign these clusters as a list to an attribute of the tweet.

Next, we delete the present tweet from stage_db and insert into commit_db.

We select only the relevant attributes of the new tweet and persist it into main_db. This is done to save space and our memory footprint, since we do not want unnecessary attributes of the tweets

Apart from maintaining clusters, we also perform other book-keeping which helps us expedite the whole clustering process. This includes (a) updating entity_db, which holds information of all the Entities in the DB (b) updating index_db, which helps us find neighborhood efficiently.

Finally, we pass back to the user all the tweets in commit_db.and delete the tweets simultaneously. This ends a successful batch session.

Even though our pipeline is straightforward and working as per our expectations, there is still room for optimization. Currently, the whole clustering process ails with too many DB queries. We plan to to minimize this overhead very soon and be able to scale up to clustering hundreds of tweets per second. That is the next challenge we’re going to work on. How we do it, I’ll let you know in my next post. Till then, enjoy Frroling and let us know what do you think.

Abhishek Vaid

@vaidabhishek

We often get this question. And many a times, the person asking the question does not exactly mean to ask what Frrole is. He is rather asking how Frrole is different than its peers - the social news apps like Flipboard, Pulse, Prismatic etc. Or even Google News for that matter. In this post, I would try to answer that question.

The ‘relevance framework’ for news has multiple axis - we think that these five are the major ones:

• Time (Is this news coming to me at the right time?)
• Speed (Did I get this news fast enough. Before others?)
• Interest (Is this news or topic of interest to me?)
• Location (Do I have news for the geography that matters to me. My city? Country? Something else?)
• Source (Where is the news coming from?)

While ‘traditional’ news products have historically focused on time and speed as the primary axis, the ‘social news’ apps have taken up interest as the primary axis, most of the time executing it via the social graph based personalization.

Frrole instead takes source and location as the primary axis. Our premise is simple - news can come from just about anybody with a phone and internet connection in today’s world, so why should the sources be limited to a few? Of course, this means a possible glut of news, and that is where location comes in as a primary filtering mechanism.

Of course, we are not the only ones who believe that this approach is a key part of the future of news. David Carr of NYTimes has talked about “moving towards a hybrid media”, Prof. Jay Rosen of NYU has talked about ‘participation of more people’, and to the best of my knowledge, both of them represent the epitome of progressive thought in either professional or academic journalism.

So having shared these thoughts, let me add - we are not suggesting making a choice between your ‘personalized’ news app and Frrole. They both are complementary, in fact. When you know what you want (your interest) or who you want it from (people you follow), use any of the personalized news apps, but when you want to know what the rest of the world is talking about, that is when you need to have the Frrole app in your pocket.

Happy frroling!

Amarpreet Kalkat

@amarpreetkalkat

So Frrole has finally launched its much awaited iPhone app. And if we go by the initial response, we must pat our backs for what have we created (Yes, we are in love with it). Anyway, first of all, we apologize for the wait that you had to go through, we could’ve released the app a few weeks earlier, but it got stuck in approvals.

Some of you have come back to us with the feedback that when you click on a story, it opens in a new browser window. This was the bone of contention that led to the delay in app release. We were ready to go live in Dec, and the app that we submitted to Apple store had the provision of opening news stories within same app window, but they raised some licensing concerns and asked us to modify the app such that news stories open in a separate browser. Moreover, we tried to register as a company with the Apple store, but there also we had to face approval delays, and later had to submit the app as an individual developer.

On the positive side, we gained some time to get feedback from friends and improve on a few design glitches. Let’s talk about what’s new in the iPhone app from the earlier Android version. First of all, the navigation is supercool – you can access all the categories (of any particular location) on the same page. Now, you can scroll up and down to see more categories and scroll left and right to see more news within same category. Second, you can now personalize your news viewing experience. Save as many location-category combinations as you want, and switch between your saved combinations just at the click of a button. The third new feature automatically detects whenever you change your location and suggests if you want to see the local news, so that you can know all that is happening around you the moment you enter a new city. All of them are very useful features I would say, thanks to Roshit@Wow Mobile for bearing with Amarpreet and his never ending requests.

So guys, lot of hard work has gone in making this app, we sincerely hope that you find the product cool enough to become a regular. Most of the people I know have become addicts, I hope you do too. Please give it a try, and whether you love it or hate it, drop in a comment to either commend or thrash us. There are no prizes for the best comment yet, but we will really appreciate your opinions and try to accommodate all your concerns at the earliest.

You can download the iPhone app from here:

https://itunes.apple.com/gb/app/frrole-the-social-newspaper/id584944153?mt=8

Keep Frroling!

Nishith Sharma

@nishithsharma