As a consequence of the social revolution we faced on the Web, news and information we daily enjoy may come from different and diverse sources which are not necessarily the traditional ones such as newspapers, either in their paper or online version, television, radio, etc. Everyone on the Web is allowed to produce and share news which can soon become viral if they follow the new media channels represented by social networks. This freedom in producing and sharing news comes with a counter-effect: the proliferation of fake news. Unfortunately, they can be very effective and may influence people and, more generally, the public opinion. We propose a combined approach of natural language and image processing that takes into account the semantics encoded within both text and images coming with news together with contextual information that may help in the classification of a news as fake or not.
Over the past few years, a number of high-profile conspiracy theories and false
stories have originated and spread on the Web. After the Boston Marathon
bombings in 2013, a large number of tweets started to claim that the
bombings were a “false flag” perpetrated by the United States government. Also,
the GamerGate controversy started as a blogpost by a jaded ex-boyfriend that
turned into a pseudo-political campaign of targeted online harassment. More
recently, the Pizzagate conspiracy – a debunked theory connecting a restaurant
and members of the US Democratic Party to a child sex ring – led to a shooting
in a Washington DC restaurant. These stories were all propagated via the use of
“alternative” news sites like Infowars and “fringe” Web communities like 4chan.
There are many reasons for the rise in alternative narratives, ranging from
libelous (e.g., to harm the image of a particular person or group), political (e.g,.
to influence voters), profit (e.g., to make money from advertising), or trolling.
The barrier of entry for such alternative news sources has been greatly reduced
by the Web and large social networks which offer the possibility to access to a
growing amount of public content that is published worldwide every day [
The problem of detecting fake news is currently widely studied in literature [
Although academic research dedicated high efforts to fake news detection
algorithms in the recent years, there is lack of solutions able to provide an
ondemand, real time, news classification. Indeed, such a system requires a wide
range of competences ranging from sociology, to combined image and Natural
Language Processing analysis, to big data management, Semantic Web, Machine
Learning, etc. [
The amount of fake news is reaching crisis proportions and it is getting worse. Among the Vs that characterize Big Data (volume, variety, velocity), we have also the challenge of data veracity. Misinformation dynamics is where the big data concept of data veracity and the problem of fake news connect. Fake news is intentional misinformation and it is also dynamic.
Leveraging artificial intelligence is definitely a must to combat fake news.
Artificial intelligence is able to learn behaviors based upon continually
improving pattern recognition, so training a system to identify fake news based upon
what sort of articles people have flagged as misinformation in the past is well
within the reach of today’s technology. Moreover, it is straightforward and very
effective to apply machine learning tools of Big Data frameworks to come up
with classification algorithms with the aim to spot fake news. In the context
of fake news, tampered images represent a crucial role in reaching virality and
allowing fake news to be shared at large scale. Although methods of multimedia
forensics are able to detect tampered images [
The proposed approach, we named NewsVallum, is expected to have high impact and to improve the state of the art on a wide range of research fields, ranging from applied artificial intelligence to the continuously disrupting field of the interaction between sociology and technology. An holistic approach is adopted, using the leading edge artificial intelligent techniques focused on image and text analysis. A joint approach to handle these two aspects that are foremost for understanding the characteristics of fake news is of paramount importance. In fact, people enjoy and share contents if they are in accordance with their vision of the world. As this subjective evaluation can be guided by both text and images, contents may be used to influence people towards a specific interpretation (this communicative process is called “framing”).
As artificial intelligence is sometimes heralded as the new industrial
revolution, a key motto of NewsVallum is to consider deep learning and reinforcement
learning as the steam engine of this revolution, whereas data can be represented
as its coal. A multimedia forensics approach is also used to make it possible to
concentrate on news manipulation strategies not only at a technological level
(i.e., using photo-editing programs), but also at a context level. In so doing, an
analytic interpretation that goes beyond the level of technological manipulation
is made, also concerning the contextualization of any given content. Another
important characteristic of NewsVallum relies on the massive exploitation of
sociological orientation and crowdsourcing. Here, the social dimension of fake news
is used as theoretical reference to understand the consequences of the social
processes concerning the main technological aspects related to the diffusion and
circulation of fake contents. Moreover, the social dimension occurs also when
the interpretation of contents is combined with multimedia forensics analysis
strategies. This tends to compensate the lack of computational social science
that typically characterizes research focused on fake news. In fact, being related
to digital platforms, current research works have too often a strong
technological bias and tend to forget the social dynamics of underlying the process.
Crowdsourcing creates a collaboration between human analysis and
computational approach, thus allowing to naturally focus on the human limits and on
the technological constraints of fake contents. The NewsVallum approach targets
two main goals. First of all, it is aimed at evaluating and assessing the reliability
of information that is spread, published and shared through web platforms on a
daily and also hourly basis. Indeed, so far the problem of automatic fake news
classification has been tackled considering either text or images. In particular,
the automatic classification of a piece of text as news has been tackled in [
NewsVallum targets both text and images together and this has also important sociological aspects and consequences for the following reasons.
The issue of fake news presents itself as the last stage of typical propaganda processes. The novelty lies in the introduction of the computational variable within the disinformation dynamics. According to the sociological theory of media construction of reality, media have a very important role in shaping our perception of the world. As long as mass media and social media keep spreading fake contents, the perception of the world could be modified accordingly, with enormous risks for democracy, security, and civil coexistence. The digital dimension of these dynamics makes these processes very complex, giving small chances to individuals in the task of finding truthful news and information. The properties of digital content such as persistence, replicability, scalability, searchability, greatly enhances the negative consequences that fake news can have on the society. This is the reason why, an innovative approach to the problem of disinformation is needed, able to combine advanced digital technologies (in particular, artificial intelligence, integrated analysis approaches, and multimedia forensics) with the goal of implementing a general framework able to give support in the task of understanding some of the social processes that underlie the current digital society. Only in this way it will be possible to develop a critical thinking positively biased by the underlying digital society. This result can be obtained by devising and implementing sophisticated technologies and approaches for fighting against a phenomenon that may have dangerous effects (on the medium and long run) on the whole society. The important and evolving requirements from the sociological point of view and the consequent technological challenges will be faced exploiting the recent revolutionary advances of AI domain. Indeed, there is evidence that modern artificial intelligence techniques (e.g. deep learning) are good at learning behaviors based on data, avoiding the influence of subjective (i.e., biased) opinions. These systems, which continuously improve the knowledge of patterns in the processed data, can be developed and trained for identifying fake news based on contents that people have flagged as misinformation in the past. Hence, this goal is well within the reach of today’s technology.
Important goals that NewsVallum aims to reach for the technologies and algorithms developed in the approach are: i) simplicity, ii) scalability, as well as iii) versatility and reusability. The first goal can be attained considering that the employment of deep learning removes the need for feature engineering, replacing complex engineering-heavy pipelines with simple, end-to-end trainable models. Scalability will be addressed taking advantage of the Moore’s law, as the proposed approaches are highly amenable to parallelization on GPUs. Versatility and reusability can be obtained as the deep learning models to be developed can be trained on additional data without restarting from scratch. This last feature is very important for the domain of fake news, in which continuous online learning is a preferred option.
Fake news detection is currently a topic of central interest at national and international level. At national level, one of the biggest Italian newspaper claims that 2017 was the year of fake news spread and even in 2018 several Italian newspapers discuss how fake news are fueling misinformation in politics, economics, and sport events. There is a specialized national police team entrusted with facing the problem, although it usually lacks of the powerful tools to deal with the problem. In the US, the spread of fake news has been seen as part of the rise of post-truth politics, in which the debate is framed by appeals to emotion disconnected from the details of policy. A study conducted by Stanford Graduate School of Education revealed difficulties of middle, high school and college students experienced in differentiating between advertisements and news articles where the information originated. The same study showed that 44% of all adults get their news from Facebook, and further investigations showed that nearly 40% of content by far-right Facebook pages and 19% of extreme left-leaning pages were false or misleading 3
The growing interest in fake news is motivated by the fact that people are typically not suited to distinguish between good information and fake news, in particular when the source of information is the Internet (and especially social media). In this context, the global nature of such an information-sharing environment allows fake news having a big impact in many fields, including politics, business, and health, at a worldwide scale. For these reasons, we expect that NewsVallum will find several practical applications that concern the verification of online contents. While providing services to detect, authenticate and check the reliability of online contents, the activities and research achievements related to NewsVallum would pursue and accelerate the technological and scientific progress in the field. Moreover, deep learning is currently considered a main research topic in the field of AI, as it creates the conditions for spreading the AI technology in various kinds of applications. NewsVallum is consistent with this trend, as it will give opportunities to young researchers for getting trained in this new and promising research topic. It is worth pointing out that the problem related with fake news, while affecting the information ecosystem, may also have negative impact on the democracy itself. In fact, fake contents may be used to generate hatred towards a social category, ruining the reputation of a political entity, and/or steering political elections with campaigns aimed at spreading false contents. Fake news is also an economic problem. Different economic stakeholders - no matter whether they are individuals, business companies or associations perform decision-making by gathering information on different topics from digital sources, such as search engines or contents conveyed by social media. This is the reason why fake news, other than being an issue for the world of information, is also a problem for the world of business. In fact, in a society in which the speed of collecting and evaluating information is a strategic variable, fake news could lead to a wrong understanding of a particular socio-political scenario, with negative economic consequences. The expected impact of NewsVallum is to provide a set of tools to be used while assessing multimedia contents with the goal of unmasking fake news, so that political and/or economic decisions can be based on a proper evaluation of the data at hand.