Twitter's new AI can find the most interesting part of all your photos


The artificial intelligence that crops your Twitter photos is getting a lot smarter.

In a new blog post, Twitter machine learning researchers Zehan Wang and Lucas Theis describe the company’s new approach to cropping your photos into preview thumbnails.

Twitter has been working on this tool for a while, but the post is the first detailed description of researchers’ methods and process. The feature is currently rolling out to all Twitter users, and aims to put an end to awkwardly cropped thumbnails.

Previously, when deciding which part of your images to display as the preview image, Twitter looked for the most prominent face. For pictures containing no faces, Twitter displayed the center of the image. If you’ve ever seen an awkward thumbnail of a cat’s neck or a white wall, that process was responsible.

Going forward, Twitter will crop using “saliency.” Saliency refers to the interestingness of a region of an image — how likely a viewer is to focus on it. The researchers cite studies showing that people tend to pay most attention to faces, text, animals, and regions of high color contrast.



The researchers have trained Twitter’s neural networks to find the most interesting parts of your photos in a very short amount of time, so you won’t notice delays while posting your photos.

Software engineers used a technique called “knowledge distillation” to train their algorithms to quickly approximate the most salient parts of your photos. While it might take a long time to make fine-tuned pixel-level predictions, Twitter’s neural networks do a speedy, more approximate version to get your photos up on time.

Engineers also used a technique called “pruning” to make sure the algorithm skips over features of your images that will take a while to investigate without yielding much benefit.

It’s a small update, but one that could dramatically improve your Twitter feed. You’ll see fewer weird, random thumbnails, and maybe even save time clicking.

How to Turn Twitter Followers into Repeat Customers [Infographic]

Twitter is a powerful tool which businesses can use to build meaningful connections with new and existing customers, brand advocates, influencers, as well as a larger, engaged audience. It’s also an increasingly important platform for providing quick, efficient customer service – of the small and medium-sized businesses that utilize Twitter, 85% agree that it’s a key medium for customer support.


How Pokemon Go Started – Infographic

Visual history of Pokemon Go creation – the timeline starts with John Hanke seeing the opportunity in interactive gaming back in 1996

Source: How Pokemon Go Started – Infographic

Since Pokemon Go launched on July 6th, 2016, it has grown faster than Tinder and Twitter. People spend twice more money on it than the gaming industry average and twice more users stay with the game than average.

People engage with Pokemons more than with Facebook. That makes it an anomaly, and, possibly, the next big thing. And this growth is only based on 8 countries where the app launched so far. How can anything grow this fast? The concept of Pokemon is not new. Neither is geolocation-based games, nor augmented reality. Reading the The Ultimate History of Video Games, you can start seeing why Pokemon Go works. The person behind the game, John Hanke who is now 49 years old, is himself surprised at how popular the game is. But not too surprised. He spent the last 20 years developing location-based interactive games. When in1996 he applied to a Berkeley MBA, in his application essay he outlined the opportunity in interactive gaming. How did he see this coming back then?

Infographic timeline of Pokemon Go history

Infographic timeline of Pokemon Go history by Anna Vital

It’s not entirely clear what Hanke did before 1996. We know that he graduated from Cross Plains High School in 1985. Then he received a bachelor’s degree from the University of Texas, Austin in 1989. He then spent some time in Myanmar (former Burma) and Washington D.C. working for the Department of State.Hanke started working on games seriously during his MBA. During the post dot-com bust years, Hanke co-founded Keyhole, a company creating geospatial visualization software. When Google acquired it, he stayed on to lead Google Earth, Google Maps, and Google StreetView development. He was also partially responsible for bringing Google Maps to the iPhone.

While many founders’ dreams run aground when their companies get acquired, Hanke not only founded a company inside Google in 2010, he was able to separate it from Google, and get Google to invest in it. How this happened is not clear. Why would Google want this to happen? Regardless, Google is one of the 3 investors in the company to this day.

The name “Niantic” comes from a Gold Rush era ship that ran aground and now lies underneath downtown San Francisco. Back when San Francisco was still called Yerba Buena, Niantic ferried tea and silk from China. But it’s fate soon changed – it was converted to a sperm whaler. It’s last voyage ironically was one of the first to bring gold seekers to San Francisco. Upon arrival, the ship was intentionally run aground and converted to a hotel. The ship is still underground in San Francisco’s financial district.

The story evokes parallels to the philosophy behind Pokemon Go – discovering the hidden treasures of one’s neighborhood, seeing something new in places you thought were completely familiar, going outside to explore.

“Question the status quo,” John Hanke advises in his alma mater promo video, “”it’s probably not how things are always gonna be.” Interestingly, the answers for Hanke seem to be the obvious things most have ignored for years – virtually reality is not a new technology, neither is geolocation. The game encourages us to explore our same old neighborhood. Perhaps even meeting the same old people we would have met otherwise. It seems we’d prefer to meet them standing next to Pokemons rather than on Tinder. The pokemons themselves have been around for 20 years. So why something so non-revolutionary could become so popular? Paul Graham, the founder of Y Combinator answered this question in his essay “Schlep Blindness”.

Exclusive Telco Research Identifies a Clear Switching Process and Social Media Recommendations (Socialyse (Havas) & Twitter)

Paris – Exclusive research from Socialyse (Havas Group’s social media solution) shows that 63% of “switchers”i will change their mobile device, and 23% will change their device and carrier. The study identifies a clear switching process including four key phases –Information Gathering, Active Research, Decision Making and Deal Hunting. It also demonstrates the important role of social media: 25% will turn to Twitter to inform their decision, and Twitter ranks among the Top 5 point of contacts for research.

Socialyse Global Managing Director Séverin Naudet comments: “The mobile phone industry is extremely dynamic. With total worldwide mobile phone shipments of just under 2 billion units, the mobile phone industry will grow by over 10% this yearii. We’ve been working closely with global telco companies, and this study really demonstrates not only how social is a powerful business solution, but how Twitter, in particular, can influence and impact the purchase decision. Social brings ROI to advertisers.”

Telco Infographic_07112016

Key Consumer Insights about Switching

The research provides keen insights into consumer attitudes and behaviors regarding switching their mobile devices and carriers:

  •   There are 3 key consumer motivations for a switch: improving current equipment, ie device and/or price; urgent needs like breakage or theft; and rewarding customer outreach at contract end-date. There were some differences by region: a higher level of motivation related to contracts in Europe and a higher level of motivation to change devices, plans or carriers in Latin America.
  •   For devices, the Top 5 criteria for a switch were (in order): Price, Technical Characteristics, Operating System, Camera Quality and Brand.
  •   For carriers, the Top 5 criteria for a switch were (in order): Network Coverage, Network Quality, Price, Internet Speed and Included Services.
  •   Despite being confused at first, most people feel positively about the change.
  •   Twitter users stay informed with news and deals available on the market, are more likely to be both experts and influencers for others seeking new devices or carriers, and are more at ease with switching.
  •   78% make their decision to switch in 1 month or less.Bruce Daisley, VP of Europe at Twitter, says: “Twitter plays a daily role in the lives of our users, and this research highlights how useful it can be in helping people across Europe make informed decisions about the telecom brands they turn to and the mobile products they buy. There are important lessons here for these brands and the industry. We found similarities in how people  approach switching from London, to Berlin, Madrid and Paris. And what’s fascinating is that Twitter can help make the process more positive. That says a lot about the platform.”

    Using Twitter to Influence Switching

    In addition to the consumer insights, the study also delivers clear recommendations on how to activate Twitter for telcos:

    •   With 25% of all respondents turning to Twitter to inform their choice of smartphone or carrier, use an everyday Twitter strategy and targeted messaging to reach users at the right time and right place.
    •   As a key touchpoint at each phase, tailor Twitter targeting to reach consumers at each distinct step in the Telco switching process: Information Gathering, Active Research, Decision Making and Deal Hunting.
    •   Consider influencer partnerships with Niche to combine the benefits of impartiality and trust with the existing power of utility and information that brands’ tweets have to drive purchase.
    •   Twitter is most impactful in the stages leading to a final decision in the switching journey. Combine an everyday strategy with TVxTwitter to drive campaign success.
    •   Twitter users tend to be more influential among their peer groups as layman experts in the telco field; Connect with and cultivate brand advocates on Twitter to grow a valuable base of earned brand coverage through outreach & advocacy.
    •   Seed promotions one month prior to the launch of a new device/plan to align with the average consumer switching cycle of 30 days.


      Research institute CSA contacted smartphone users aged 18-50 who have or will change their mobile device and/or plan in the past/next 3 months, across the UK, France, Germany, Spain, Brazil and Mexico. The study included two phases: 1) a qualitative analysis including focus groups and one-on- one interviews, and 2) an in-depth qualitative online survey.


    About Socialyse

    Launched in 2013, Socialyse is the social media solution of Havas Group. Socialyse provides a unique integrated and synchronized social media offer, including strategy, content, media and analytics. Innovative tools include the Social Rating Point and the powerful Socialyse Newsroom. With specialized talent and best-in-class technology, Socialyse guarantees both performance and prices. With over 720 social media experts based in 38 local offices and operating in 80 countries, Socialyse combines the agility of a startup and the strength of a powerful global network.


    Robert Fridovich
    Tel +33 146933715
    Mob +33 632063816

Social Rating Point Record pour Suède – Belgique: 11 ! TV et réseaux sociaux en synergie. (Source: CSA – Havas)

Au lendemain de la victoire des Belges et de leur accession aux 1/8 de finales, les équipes de CSA Belgium (Consumer Science and Analytics – Havas Group) ont analysé l’engagement des Belges sur les réseaux sociaux et en particulier sur Twitter.

En croisant cette donnée avec la couverture TV réalisée pour chaque match, CSA a donc pu obtenir l’index du Social Rating Point* des 3 matchs.

Notons au passage que 80% des 15-54 étaient présents devant leur écran pour suivre les matchs de poule.

#belita: 37.000 tweets pour une audience de 1,2Mio – SRP = 3

#belirl: 75.000 tweets pour une audience de 935K – SRP = 8

#swebel: 129.000 tweets pour une audience 1,2Mio = SRP = 11

Sans surprise, ce sont les hashtags officiels des matchs qui récoltent le plus franc succès – 70% du volume.


Au niveau des joueurs, Romelu Lukaku est le joueur des Diables Rouges qui génère le plus de tweets, il faut bien l’avouer, desservi par sa prestation du premier match contre l’Italie.

Les équipes de CSA se sont également intéressées à la visibilité des Diables Rouges au niveau mondial. Résultat: 1 twittos sur 3 a eu minimum un tweet lié aux prestations des Diables dans son feed. Ce qui représente une audience globalisée de 90 millions d’utilisateurs! Belle visibilité donc pour notre équipe nationale.


*SRP = Le SRP est calculés en divisant le volume de mentions d’une émission par son reach. Avec une base de données de plusieurs milliers d’émissions de télévision – chaînes généralistes et spécialisées -, ils permettent aujourd’hui de mieux comprendre les interactions sociales pour transformer positivement l’engagement du public en business potentiel.

Twitter’s top brands for customer service | Marketing Week

on 8 Apr 2016

Brands are increasingly using Twitter as a customer service tool but not all are getting it right, according to a new study, which shows that retailers are the worst performers while the financial sector is leading by example.

Source: Twitter’s top brands for customer service | Marketing Week


In today’s social media age, Twitter-based customer service is expected to be speedy. But according to mystery shopper research by BDRC Continental, many brands struggle to keep pace.

Established retailers are among the worst brands for answering customer queries on Twitter, while high street banks and credit card brands are the top performers.

Silence is costly

Fashion retailer French Connection, which saw sales in 2015 fall 8% to £164.2m, comes bottom of the pile in the research, taking an average of 39 hours and 22 minutes to respondto customers on Twitter.

“The longer a brand waits, the bigger the risk that a customer will decide to go and shop somewhere else,” says Tim Barber, director at BDRC Continental. “It is 2016 and people expect a rapid response. If they don’t get it, you could lose them forever. French Connection could be unintentionally sending that waiting customer to Topshop.”

With an overall score of 10% across several metrics including customer satisfaction and speed, French Connection is the worst-performing UK brand according to the study, but it was not the only guilty high street retailer. The brand is joined by high street giants including Starbucks, Waterstones, Vodafone and McDonald’s, which are among the 10 worst performers.

As part of the research, more than 9,000 tweets were sent to 395 high street brands in 32 market sectors. Each of the brands’ responses was ranked by the quality of their answer, whether additional information was offered, the friendliness of tone and if a reasonable effort was put into the response. All brands were sent the same 25 queries by 25 different users to allow for benchmarking of the responses.


Barber says the research produced its fair share of horror stories. “The worst thing a brand can do is respond even when they haven’t properly read a question or use automation to just cut and paste answers.”

He reveals: “One of our shoppers asked a payments brand if there was a charge to pay by American Express, and it bizarrely responded: ‘We take MasterCard, Visa and American Express’.

“Another shopper asked a popular high street restaurant if they could bring their own wine to a meal or if there was a charge. The response was an aggressive: ’No, absolutely not. If you do, you will be instantly removed from the premises’.”

Finance industry are top tweeters

High street retailers could learn a thing or two from the financial sector, however. According to the report, Nationwide has the best customer service on Twitter with an overall score of 88%. In fact, seven of the top 10 best performers are either credit card or banking brands.

The financial industry has had its fair share of PR disasters since the 2008 banking crisis andBarber says this works in its favour. “The banks, credit cards and rail companies are all used to being attacked regularly in the media. As a result, they have huge teams in place to manage their social media and PR. If they can make a customer happy on Twitter, it’s an easy PR win.”

At four minutes and 25 seconds, Virgin Trains has the quickest average response time. However, Barber says speed is not everything and that there is still a trend of “quick answers that don’t really answer a consumer’s question”.

The rise of online forums also means that the public can find answers to most queries through the simplicity of a Google search, suggesting brands that cannot do a good job of customer service on Twitter should not do it at all.

“Apple doesn’t answer any tweets, so it isn’t necessarily imperative that you go down the route of using Twitter as a customer service tool,” says Barber. “But if you are on the platform, you should invest in doing the job properly as it will pay for itself in the end. If you’re on Twitter and responding to customers poorly, it will do more damage than good.”


Ultimately, Barber says brands must be consistent if they are to achieve success through social media customer service. He suggest brands such as Holland & Barrett and Waitrose, which achieve scores of 77% and 75% respectively, do well to balance both helpfulness and personality.

Barber advises: “You need to be consistent with every single tweet and treat everybody like an individual. You have to have the infrastructure in place to respond to every single tweet – that’s important.

“You don’t pick and choose who you speak to when someone walks into a store – you aim to speak to everyone, so why do the opposite on Twitter? Treat every Twitter user like a shopper inside your flagship store. Remain consistently helpful and you’ll inevitably improve the sentiment towards your brand.”

More than half of global internet users are using messaging apps on a daily basis | TNS Global

New forms of social platforms present a new set of challenges for brands – and they can’t afford to duck them

Source: Our newly complicated social lives | TNS Global

For the best part of a decade, our online social lives have been defined by a handful of giant networks: Facebook, Twitter and (for professional and high-net-worth audiences) LinkedIn. Marketing strategies have therefore focused on capturing attention and driving engagement through these channels. With their mass reach and openness to advertising, they had appeared to make the task of engaging audiences on social media relatively straightforward. But things are changing – and changing fast. As far as marketers are concerned, our social lives are becoming a lot more complicated.

It’s not that the social giants are in decline. Their numbers of active users continue to increase, with 30% of global internet users heading onto Facebook every day. But they no longer define our digital social lives in the way that they once did. Tweet thisThe average 16-24-year-old now uses at least five different social platforms each week. And many of the interactions that matter most to them take place on new types of platform, with less obvious roles for brands. This represents a challenge for marketers, but it’s also an important, and very timely, opportunity.

The daily use of mainstream social media is rising at 6% a year – but use of Instant Messaging (IM) platforms is increasing at double that rate. More than half of global internet users are using messaging apps on a daily basis – and this is pushing platforms such as WhatsApp, WeChat, Viber, Snapchat and LINE towards the centre of social experiences. Messaging use may lag behind in the US and UK (35% and 39% respectively), but it dominates in emerging and fast-growth markets from Brazil (73%) to Malaysia (77%) and China (69%). And it’s not just IM that is expanding the social media universe. From Instagram to Vine, people are engaging on a far broader range of platforms that reflect their different needs and interests through very different social experiences.

It’s significant that Facebook itself has moved swiftly into these new social spaces: WhatsApp (owned by Facebook since late 2014) and Facebook Messenger are currently the two largest messaging platforms on earth. Facebook owns Instagram as well. The world’s largest social media player recognises that people are no longer satisfied with expressing themselves in just one way, on one type of network.

Look closely at the IM phenomenon and you’ll find a broad range of user experiences from stripped down to content-rich, but all have the same fundamental appeal at their core. IM enables real-time interaction within small, specific groups that can form, break up and re-form endlessly. Conversations are free to fade over time (the importance of which explains the intense debate about just how long Snapchat conversations are stored for). IM is more intimate, more responsive and less permanent. And it reflects more intuitively the way people have traditionally interacted with one another.


The return of nuanced private lives image

The return of nuanced private lives

To marketers, the sudden fragmentation of the social media landscape can seem disruptive and surprising. But in the context of human relationships, the shift isn’t really that surprising at all. If anything, it’s the last 10 years that have been abnormal.

In the offline world, there are certain, very specific situations in which being included in a large crowd is an important part of the experience: shopping in a marketplace, watching a football match, engaging in a pilgrimage or, on a smaller scale, enjoying a large dinner with extended family. But these situations are relatively rare. The vast majority of our lives aren’t spent shouting personal messages over the heads of very large crowds of people; they are lived more privately, built on interactions with far smaller groups that we shuffle between according to circumstance and need.


Living out more of our life online – but on different terms

The shift towards IM suggests that people want to move more of these offline experiences onto digital platforms; they are just not comfortable moving them onto the likes of Facebook. Messaging apps provide them with opportunities to reflect a more nuanced social reality in the online space. And as those messaging platforms build richer functionality around core messaging, there are more compelling reasons to do so. Users of LINE, which first launched as an emergency response to the Japanese earthquake of 2011, can now play games together, generate and share content, stream music and watch TV – all of which can be plugged seamlessly into their interactions.

More consumers carrying out more of their social lives online ought to represent a significant opportunity for marketers. However, there’s a challenge. The way that brands interact with consumers on Facebook, for example, is partly why large-scale open networks are deemed inappropriate for the more private and personal aspects of people’s lives. Facebook feels like the virtual equivalent of the marketplace, an environment in which we must be happy to be distracted by a host of different voices and different propositions calling for our attention; messaging platforms do not.


Time for a holistic social media strategy

There is still an important role for the marketplace in people’s lives – a space where we seek out new forms of stimulation and can quickly gain a wider sense of what’s going on in the world. Within our newly complicated social lives, the giant, open networks that provide this experience are the natural home for scalable, awareness-building, amplification and engagement campaigns. However, when connected consumers move onto IM platforms, they no longer welcome distractions in the same way. WhatsApp’s ‘no ads’ policy is a significant part of its appeal – and brands must proceed with caution even on platforms that are more open to marketing. People resent unsolicited brand messaging popping into the same space that they use to chat confidentially to close friends and family. When they do invite brands to engage with them, they expect them to behave in a manner that reflects this context. This means that marketers must play by a new set of rules if they want to thrive. We’ve been accustomed to seeing social media as one element within an integrated marketing strategy. Now we need to take a more granular view of the social media space itself – and start adopting integrated social media strategies that reflect the roles that different platforms play.

Time for a holistic social media strategy


Getting in touch with humanity

The new rules are defined by the nature of messaging platforms themselves. These platforms deal in human experiences that are intimate, intuitive and relevant to the moment. Success for brands will similarly depend on a more human approach: facilitating dialogue and connection, replying to comments in a way that is worthy of that dialogue, caring about the moments that matter to the audience, and delivering content that is genuinely relevant to those moments. The most successful brands know that such relevance can come through providing tangible value that enhances or simplifies life – but also through the emotional connection that is formed when they align themselves with the right format and tone of voice for a platform.

Taco Bell’s use of Snapchat excels in establishing such a human connection. It embraces risk, experiments with different approaches – and earns respect from consumers for avoiding a bland, corporate tone of voice. Its choice of influencers to enlist for the brand has been appropriate as well: focusing on YouTube and other social media celebrities who have credibility on the platform, rather than conventional star power. The six-second DIY videos posted on Vine by the US home improvement store Lowe’s similarly mixed tangible value with emotional engagement by working within the format of the platform. The result? Lowe’s ninety Vine videos delivered over 33 million views.


Sharing counts more when it’s personal

Tweet thisAs social platforms become more personal, the credibility added when content is shared becomes all the more valuable. WeChat Moments, which builds a broader social media experience onto WeChat’s IM platform, provides a powerful amplification opportunity for brands, with any content that a user engages with hugely more likely to appear in the feeds of their friends. The power of personalisation can also be seen in the influence of Instagram, which has grown in value as an earned media opportunity for brands. Instagram provides another environment where sharing amongst a more select network delivers powerful benefits. It was recently claimed (in research commissioned by an Instagram influencer app, Takumi) that 68% of 18-24-year-olds were more likely to purchase an item if someone they followed on Instagram shared it.


Why a more complex social world has more opportunity

Tweet thisFinding the right strategy for more intimate social platforms will become increasingly important as those platforms take on a broader range of roles in consumer lives. Users of WeChat can already browse and buy products (with 10 million WeChat stores opening in the last year alone), order taxis, apply for loans and keep track of their fitness. Facebook Messenger’s AI-powered virtual assistant, launched in September, prompts users with suggestions for where to eat, where to visit and what to buy. As social media becomes more varied, more versatile and more personalised, consumers are far more likely to trust the right social channel to guide their choices. As these new social channels become more sophisticated and seamless, people could increasingly make the journey from awareness through to purchase without ever leaving the platform they are on. Making themselves a part of this environment will require brands to invest in customised approaches, explore more partnerships, be more human and less guarded in their approach, and take more risks. But all the indications are that such effort will be worth it. Our online social lives may be becoming more complex – but adapting to that complexity will be increasingly rewarding.


Anjali Puriis Global Head of TNS Qualitative. Anjali is responsible for developing TNS’s qualitative offer, providing clients with cross-cultural insights, and leading new thinking, particularly in the areas of consumer choices, behaviour change and social media.

Zoë Lawrenceis Marketing & Communications Director, TNS Asia Pacific. A member of the APAC digital board, Zoe has been involved in shaping TNS’s thought leadership around the connected consumer since 2010.

Twitter est-il vraiment la « boule de cristal » des marchés financiers ? – Rue89 – L’Obs

Twitter est-il vraiment la « boule de cristal » des marchés financiers ? – Rue89 – L’Obs.

C’est un article des Echos qui a provoqué le courroux de Thomas Renault, doctorant en économie, qui anime le blog Captain Economics.

« Comment Twitter est devenu la boule de cristal des marchés financiers, » annonce tout de go, le 31 juillet, le quotidien de l’économie. S’ensuit une mignonne infographie sur les « tweets qui ont fait réagir les marchés » – comme celui posté par le compte piraté d’Associated Press annonçant une explosion à la Maison-Blanche – et un papier qui tend à montrer que le réseau social est « devenu la miss météo des marchés ».

A l’appui de la démonstration, Les Echos citent une étude de Bollen, Mao et Zeng, chercheurs à l’université de l’Indiana et de Manchester, qui, en 2010, ont montré que les données de Twitter pouvaient prédire l’évolution du Dow Jones avec 87,6% de fiabilité [PDF]. Autre étude, même effets : des économistes de l’université de Californie ont trouvé une « corrélation importante entre le nombre de transactions d’un titre et le nombre de “composants raccordés” – c’est à dire le nombre de posts liés à des sujets distincts concernant la même entreprise » [PDF].

Bref, tout cela a pour but de montrer qu’il est possible de « sentir » les états d’âme – irrationnels ou non – des investisseurs de toute farine à partir de l’analyse d’une masse de signes papillonnants (les tweets). Tout cela semble validé par une étude récente de la Banque centrale européenne (BCE) [PDF].

Sauf que.

Méthodo ?

Thomas Renault, qui prépare une thèse sur le sujet et dont les posts sont relayés par le très libéral site Contretemps ainsi que par divers organes de presse, appelle à la prudence.

En préalable, il donne un petit cadre conceptuel à sa critique : deux visions pourraient expliquer le rôle de Twitter sur l’évolution des marchés. La première se base sur la théorie informationnelle : l’idée que l’information publiée sur Twitter n’est pas encore intégrée dans la formation des prix. Elle est fondamentalement nouvelle. Ainsi, l’arrivée de Twitter modifie de manière permanente la formation de ces prix. La seconde se rapproche de la théorie « sentimentale » : le prix d’un actif peut dévier de son niveau d’équilibre en fonction du sentiment d’investisseurs « naïfs ». Twitter sert justement à mesurer ces craintes ou enthousiasmes.

Bon. Mais cela ne remet pas en question la pertinence des études citées plus haut. Cette critique constitue la deuxième partie de la note de Thomas Renault.

Argument numéro 1 : l’étude de Bollen, Mao et Zeng, qui prédit avec 87,6% de fiabilité l’évolution du Dow Jones n’est pas si béton que cela :

« La précision du modèle est testée “out-of-the-sample” sur une période allant du 1er décembre 2008 au 19 décembre 2008, soit 15 jours de trading ! 15 jours, en données daily (donc une étude sur 15 points) ! De plus cette étude ne s’intéresse pas au rendement d’une stratégie de trading basée sur Twitter, mais simplement à une prévision de la direction du marché (est-ce que le marché va monter ou bien est-ce qu’il va baisser). Pour finir, 6 modèles différents sont testés et le résultat mis en avant est celui du meilleur modèle, ce qui a tendance à clairement biaiser les résultats (en testant un grand nombre de stratégies purement aléatoires sur un nombre de points limités, la meilleur stratégie aura statistiquement une précision proche de 100%…). »

Argument numéro 2 : l’étude de la BCE n’est en réalité pas une étude de la BCE. C’est même précisé au début : « Ce papier ne représente pas les vues » etc. Il est signé par deux des auteurs (Bollen et Mao) de l’étude précédente.

Argument numéro 3 : corrélation n’est pas causalité.

« Dans l’article des Echos, il est par exemple écrit “A l’époque, Bloomberg en avait également profité pour sauter le pas en ajoutant les messages Twitter à son offre de services financiers quotidiens. Dans les semaines qui suivent, les ‘ flashs crashs ’ s’enchaînent.” Alors oui, le fait que des algos de trading haute-fréquence puissent analyser en temps réel le contenu des tweets (et tout autre flux d’information – articles, blogs, forums…) pour prendre des positions peut en effet déstabiliser les marchés en augmentant la volatilité… Le Captain’ n’a rien contre cette hypothèse, mais comme toute hypothèse, il va falloir prouver cela empiriquement en “contrôlant” le tout pour considérer l’ensemble des autres explications possibles. »

Conclusion de Thomas Renault, qui cite à son tour des études moins tranchéessur la question :

« Les résultats empiriques sont pour le moment très mitigés. […] Twitter n’est clairement pas une “boule de cristal” permettant de prévoir les marchés, mais plutôt “un bon miroir” reflétant la situation présente. »