Social Networking Services and Technology Development
Social networking services have undergone a lot of changes ever since the concept of Web 2.0 has been proposed, promoting sharing, participation and openness. The era where one-sided information sharing through a website was the mainstream now has changed into the era where communication and collaboration has become pivotal. Social networking services that enable people to learn and share personalized information in this social trend have become the new normal.
Until just recently, people using Facebook or Twitter were accepted as early adopters, but those who have only recently joined these services should be considered as the late majority.
Usually, when a service reaches maturity in its lifecycle, there will be other services aiming at a niche market. The Social networking services that we are familiar with— Instagram, LinkedIn, etc., aside from those mentioned above— are actively used as well. It’s not difficult to find services provided for a specific group such as ResearchGate (researchers' social network service). One of the factors that enable the emergence of new forms of social networking services that meet these diverse needs is the development of technologies.
The writing technology related to the initial social networking services didn’t require highly advanced technologies. On top of the infrastructure, such as a network or server, the technology of creating websites or apps, plus the friend recommendation technology based on feed technology and profile analysis were all that’s needed.
However, in recent years, services have become more intelligent and require higher levels of technologies. The top 10 strategic technologies Gartner Group announced in 2011 included social analytics. Social analytics is a technique for measuring, analyzing, and interpreting the interaction or association between people, events, and ideas, which can be used not only in physical spaces such as work or social gatherings,
but also in online contacts and relationships.
Specifically, it is a concept that includes professional analysis techniques, encompassing social search, social filtering, social network analysis, emotional analysis, and social media analysis. In particular, social search is a technique used to search for information produced by social networking services rather than general website analysis such as Google.
Since social data is mostly unrefined data and related information is added rather than simple one-way information offered on a web site, the search method and contents must be different, and a social search technology, taking this factor into consideration, is currently being developed. In addition, location-based technology, text mining technology, and AI are technical elements that enhance the value of social network services. In this article, we will discuss how AI can be deployed in social network contents and social networking service analysis.
Social Network Contents Creation & AI
What matters most in social networking services are the contents. The reason most people use social networking services is because they want to know the current status of people connected with them, but communicating with others and getting useful information are their important purposes as well.
In the end, social networks are also the sources of information on which network externality is based. However, as the phenomenon of the alienated class that could not participate in the Internet society in the past is called the digital divide, a class that cannot be assimilated into the social networking services has newly emerged. It is not uncommon to find that many people are afraid of posting their thoughts on social networking sites because of other people’s negative reactions.
AI can solve these people's challenges.
Post intelligence, recently developed by the US Company, helps people who do not know what content to put on social networking sites by providing relevant contents through AI.
AI that could analyze the articles posted on those sites by a user, the habits, the speech, and even the tendency of the follower, and could suggest the contents to be posted for the user are developed.
It recommends different ways to easily use social networking by suggesting contents or topics to post for attracting people’s attention and ways to boost the number of followers.
Technically, Post Intelligence uses deep learning to analyze tweets of users and followers, learn what followers like, recommend engaging posts by learning and predicting contents followers might enjoy, and predict how well their post will do.
To do this, AI needs to learn a vast amount of social network information.
Not only the content of the posts made by the users, but also the reviews of their followers, their profiles, and so on, must be analyzed to find the patterns and be used in order to create new contents. Considering the amount of data and the speed of learning, people are not suitable for these processes, unlike AI which has the strengths to pull those off.
People’s interest in services, recommending content with AI on user reviews, ratings, opinions, and information in social media is ever increasing day by day.
For example, take a movie recommendation service. Netflix, the largest online video service provider in the world, is using its sophisticated recommendation system to increase the satisfaction of its subscribers as it can freely stream a vast amount of movies online.
Movie recommendation techniques are basically implemented by film-related data and machine learning. Quantitative data pertaining to movies are objective data such as movie genres, screening times, and box-office performances, and qualitative data are subjective data including the evaluations of viewers such as movie atmosphere and story complexity.
Especially majority of subjective data are exposed to social networking and analyzed using text mining techniques.
Through the collected data, movies are broken down into long lists of attributes to match these elements to a viewer’s preferences. These values are stored in a database, and processed through machine learning to recommend movies that match the preference of a specific viewer.
In recent years, techniques for recommending movies suitable for the viewers with AI have been applied by collecting ratings and information about movies on social media and taking time, place, and social contexts into account.
Recommendations leveraging AI can be extended to social shopping areas. Of course, this is also the technology that can be used across the entire online shopping. In the distribution market, activities of buying and selling goods are repeated every day, which allows large amounts of data to be accumulated. Based on this characteristic, the distribution industry recommends appropriate products to customers.
Domestic department stores or online shopping malls are already utilizing their own AI to identify customers’ favorite brands and send relevant information to mobile apps, or enable customers to find products easily in a few conversations with a Chabot. On the other hand, social shopping provides more intelligent services by using text information of social networking services or log information collected from social graph.
Social Networking Services Analytics & AI
It is essential to analyze the relationship of service users and network characteristics in a situation where social networking services are extended not only to daily communication but also to acquisition of information.
Social networking services’ relationship with users is a good condition for social analysis to analyze networks consisting of nodes and links. Social analysis can measure relationships among people, intimacy, grouping, and connection strength.
Specially it calculates indicators such as betweenness centrality and closeness centrality, and suggests what role each node occupies in the network. In social networking services, these centralities can be used to determine how many users are directly connected to each other or the extent to which the relationship between users is mediated.
That is, it can extract who the key user is in the social networking service. Furthermore, social analysis, combined with various theories, provides insights into the special phenomena that occur in the network.
Theories— social influence network theory, social capital theory and theory of strength of weak ties— can be adopted to analyze social networking services, for instance, theory of strength of weak ties indicates that acquaintances are more likely to provide new information; indeed, the more weak ties we have, the more connected to the world we are and are more likely to receive important information.
The types of information that can be utilized to perform social analysis in social networking services can largely be divided into follow, reply, and retweet (Twitter). When you want to make friends with a specific person online or want to see his / her posts, you follow them; make a comment on their posts or retweet, which means more active interaction.
Analyzing these relationships make it possible to evaluate the influence or propagation power of a user, and to track how specific posts spread or to estimate the spread of opinions in the future.
In the first article of this series, I have covered Opinion Mining, but it is not easy to accurately analyze opinions with just the text. In addition to emotional analysis results from the text, obtaining much more meaningful results is achievable if we can analyze the opinion propagation structure based on social networks.
In other words, we can get more information when the analysis is focused on people rather than text. Analyzing only the keywords provided by text, recognizing messages spread, and focusing on the relationships surrounding the text posted can illustrate the information about whose words are reliable, what messages are more powerful and how the messages spread.
In social analysis, the part where artificial intelligence can be used is the link prediction of the nodes in the network. Namely, an analysis on the relationship between the nodes disconnected enables AI to connect the nodes. Link prediction is widely used in broad domains.
In internet / web science, link prediction is used to automatically create hyperlinks or to predict website hyperlinks, and is useful for constructing recommendation systems in e-commerce. This method can also find potential links by exploiting the number of common neighbors or by finding the shortest path distance between unconnected nodes.
These days as the network grows and high volumes of data accumulate, large number of AI techniques which learn whether the link exists between existing nodes and apply the results to link prediction have been employed.
Supervised learning algorithms, including neural network and Support Vector Machine (SVM), learn the characteristics of the relationship between the nodes where the link occurs, and based on this, recommend nodes that can be connected when new nodes appear, or draw a pair of nodes that are not connected even though they must be connected between existing nodes. From the perspective of social networking services, it is possible to predict "friends" by predicting association among users and to predict the dynamic evolution of the entire network to which they belong.
A list of 'People You May Know' that we come across when we access social networking sites can be created through this AI-based link prediction.
We sometimes find friends whom we lost contact with for a long time through this link prediction, which surprises us constantly with ever-evolving technologies. The success or failure of this method depends on the definition of 'attributes' that learn and analyze the relationship between nodes. If properties that describe social networking relationships are set appropriately, AI techniques will perform link prediction successfully, and as a result will become a core function of social networks.
Social Big Data Analytics & Future of AI
Big data analytics being mass-produced in social networking services has done things we could’ve never imagined in the past, but it also has its limitations.
First of all, even though they are big data, gathering and analyzing useful social data are not as easy as you may think. The language we use in general and the language we use online are often different; the fact that production / circulation of false news and articles are widely prevalent makes it more difficult to reflect and analyze these data.
The key to social big data analytics is not about being big but about being complex. The analytics method focusing on the vastness of data is hard to give valuable information. It should be able to unravel the complexity of social data for decision making.
This is the strength of AI and will determine the direction of future advancement.
In a three-game Go match between AlphaGo 2.0 and Ke Jie not long ago, AI again attested its excellence in learning ability.
Not only does it learn numerous strategies and moves but it also learns the outcome of the match of the day before, playing more advanced game than before. Through its continuous learning, the limitations of social data analytics we are currently experiencing can be overcome.
Analyzing the complexity of social networks can present intelligent information ranging from analysis on the content of text to link relation analysis.
"The AlphaGo 2.0 is an advanced version with self-learning after learning the basic strategies and moves of Go game," said Demis Hassabis, CEO of Google DeepMind. It has evolved into the stages of creating some truly beautiful and innovative moves that were never recorded before.
This is referred as the link prediction in social networks based on the AI covered above. Soon, we will get to meet smarter social analytics thanks to AI.
Through constant learning, it will be able to interpret the user's language according to the context, filter out false news and articles, and analyze and predict complex relationships among users.
To do so, AI algorithms should continue to evolve, but the nature of social network data should be well understood and reflected.
These efforts will help vitalize the development of services based on more intelligent social big data analytics.
▶ The contents are protected by copyrights laws and the copyrights are owned by the creator.
▶ Re-use or reproduction as well as commercial use of the contents without prior consent is strictly prohibited.
Professor Yoon Byeong-un received his bachelor's, master's, and doctoral degrees in Industrial Engineering from Seoul National University. After completing postdoc at Centre for Technology Management (CTM) of Cambridge University in England, he is now a professor of the Department of Industrial Systems Engineering at Dongguk University. His areas of study include technology forecasting, technology roadmap, patent analysis, artificial intelligence, big data analysis, and technology-human society convergence. In recent years, Prof. Yoon has been making efforts to establish and spread the concept of technology intelligence.