Menu
Home
Log in / Register


 
Home arrow Computer Science arrow Social Informatics
< Prev   CONTENTS   Next >

2 Methodology

(Vkontakte(VK)) is a social networking site (SNS) with over 200 million accounts. It offers a platform for communities where users can discuss topics of their interests. Users can communicate either in a continuous message flow on the group 'wall', or in discussions sections that have their own specific subtopics.

We collected text data of apartment buildings networking groups by using VK Application Programming Interface (API). We looked for groups related to neighborhood communities in apartment buildings for all streets of SaintPetersburg. Out of total 2232 groups, we have filtered out spam and irrelevant search results, which left us with 420 groups for further analysis, 199 of which were open for data collecting. Groups were distributed evenly and covered all districts of the city, apart of industrial areas.

For exploratory analysis of the content of group discussions, we treated messages from each group as a single document and used Principal Component Analysis on a term-document matrix.

For more detailed analysis of topics in the VK groups, we used Latent Dirichlet Allocation (LDA)[1]. We aggregated all posts and comments for each group into a single document. After that, we applied lemmatization algorithm[4] and used resulting text to generate topic models. Topic models had to identify 100 topics, which contained significantly interconnected words in text documents.

After that, we clusterized the document-topic matrix to find groups with a similar topical structure and performed further analysis of each cluster.

3 Results

Daly, Dahlem and Quercia[2] found that activity of users in city forums and topic structure differ between apartment buildings, street and area groups. In our research we found that on a street or district level there were very few groups dedicated to urban topics.

Our three-factor PCA model has shown that discussions of car parking and administrative topics were two major factors responsible for the best part of diversity in group discussions. On the other hand, casual communication based on socialization between residents was not playing any visible role in the group interactions.

Further exploration of group clusters based on similarity of LDA generated topics (Table 1) has allowed us to discover some patterns of cooperation.

Table 1. Group clusters and topics

The administrative cluster includes groups where residents discuss work of HOA or management companies, state of their buildings and their surrounding areas. We have identified two types of such groups: created by residents and created by board of HOA. Communities hosted by the board of HOA are filled with spam messages and have low social activity. In contrast, groups created by residents show significantly higher social activity. In these groups, people look for ways to resolve their problems with improvement of living conditions and to organize collective actions for this purpose. They share photos of leaking ceilings, dirty staircases, broken windows etc., criticize management companies or HOA, and discuss ways of resolving their problems.

We also found a set of seven groups dedicated to protection of public locations (squares and yards). Five of these groups are located in one district of St.Petersburg. Members of these groups try to prevent in-fill constructions in the surrounding area. It should be noted that a number of people belong to several groups at once (30 individuals out of 570). We suggest that these people represent a group of local activists whose relationships are built on mutual help and support.

In these groups, activists post information about meetings with local authorities and organize or plan rallies against actions of building developers. Discussions are usually moderated by administrators who are also the most active posters. A large number of reposts from walls of other protests groups and absence of spam messages (which are deleted by administrators) are two main features of these groups. Communication between group members is often supported by snapshots made during rallies or legal acts pointing at illegal sealing constructions. The life cycle of these groups depends on the offline situation in general; we suggest that death of communication in virtual group is triggered by the end of fighting.

 
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
 
Subjects
Accounting
Business & Finance
Communication
Computer Science
Economics
Education
Engineering
Environment
Geography
Health
History
Language & Literature
Law
Management
Marketing
Philosophy
Political science
Psychology
Religion
Sociology
Travel