India is going digital in a big way; from banking to manufacturing to agriculture, each field is seeing the penetration of technology. Police organizations also have started using technology for effective policing. Most police organisations now have an official website, a Facebook page and a Twitter handle. Police not only use these new media services to showcase their organisation but also to interact with citizens very regularly. Police posts on Facebook or tweets on Twitter include a variety of topics ranging from traffic advisories, to awareness creation to bragging about their achievements. Similarly, the growing technology savvy population of India is using these mediums to share their grievances, concerns, etc. with the police. With a handful of police officers serving 1.25 billion people, it is no surprise that a lot of posts/tweets by the citizens go unnoticed by the police. Even features like tagging police commissioners and police accounts do not always yield the expected response, causing a sense of resentment. The police too find themselves helpless given the multitude of things.
With our continued interest in empowering police organizations with technology which can help them in their day-to-day activities, we have been working in the space of online social media and policing for some time now. For our research publications in this space, please visit here. For effective communication between the citizens and police, it is necessary for the police to understand the vast amount of content generated on their social media accounts. In this direction, we started thinking about how to break up the content into important versus unimportant, urgent versus non-urgent, etc. Our main aim in this research was to help police identify ‘serviceable’ content which can be served quickly and efficiently. Requests to which police should respond, evaluate or take action are considered as serviceable requests.
We analyzed 85 official Facebook pages of police organizations in India and studied the nature of posts that citizens share on police Facebook pages. Not all posts require the same amount of attention from the police, there are some cases where immediate action needs to be taken while some can wait. Based on this analysis, we came up with six textual attributes that can identify serviceable posts; posts that need some kind of police response. We find such posts are marked by high negative emotions, more factual, and objective content such as location and time of incidences.
We identify four types of response that citizens may get on their posts:
(a) Forward: Posts which had enough information and could be forwarded to appropriate authorities for action. For instance, a resident posted, Date : 4/11/2015 (Wednesday), Time : 10:17 pm, Number : [withheld], Location : [withheld], Violations : Crossing line by way too much obstructing the vehicles which were coming from [withheld] entrance later he jumped the signal ……..
(b) Give Solution: Posts mostly included queries by residents to police that could be answered without any detail; resident asks, Admin !! Can U Explain to Me How Two Challans On Same Date Same Time in Just 5 Minutes Gap !! How Its Possible ?? Any Thing Wrong ??
(c) Acknowledge with thanks: Posts to which the police wrote “thanks for sharing the information” or “thanks for the appreciation.” For instance, resident remarks, Chennai City Traffic Police a humble salute from a fellow Chennaiite for the commendable job in such rains!!
(d) Need more details: In these resident’s posts, police inquired more details so that action could be taken, e.g., a resident asks, Cops driving wrong side [of road] near XXX hotel .. what action will be taken against them ? This post lacks information such as time and date when the incident happened.
To enhance response to serviceable posts, we propose a request – response identification framework. The approach followed in the paper is shown below:
Understanding Requests from Citizens:
Residents often use different language styles in posts while expressing their concerns and asking queries to police. Our approach includes following six category of features to characterize serviceable posts:Emotional Attributes,Cognitive and Interpersonal Attributes, Linguistic Attributes, Question Asking Attributes, Entity-Based Attributes, and Topical Attributes. These include the both handcrafted features and LDA / NMF based features that help automatically discover the latent dimensions and induce semantic features in our data.
Our analysis shows some intriguing results:
Serviceable requests show significantly higher value of negative emotional states i.e. “anger” (+15.38%), “disgust” (+47.8%), “fear” (+60%), and “sadness” (+10%) in comparison to non-serviceable requests. Most frequent topic is includes queries / question posed to police (Complaints represents complaints against cops in- correct decisions).
Comparing serviceable sub-types, we observe that 93.10% posts in Thanks sub-type did not receive a response from police. Posts in Forward sub-type received the maximum number of responses from police (63.6%, 182 posts). Table 1 below summarizes the number of posts that did not receive police responses.
Table 1: Number of posts that received responses (N of Events) and censored event showing posts that did not get response from the police.
Automated Classifier for Serviceability:
Our work explores a series of statistical models to predict serviceable posts and its different types. The model makes use of the content based measures – emotions, cognitive attributes, linguistic, question posed, entity and topical attributes. We explore five different classification algorithms – Random Forest (RF), Logistic Regression (LR), Decision Trees (DT), Adaptive Boosted Decision Trees (ADT), and Gradient Boosting Classifier (GBC) using balanced class weights. Table 2 below reports the performance of different algorithms to correctly identify serviceable posts.
Table 2: Mean Performance after 10-fold CV of different algorithms to correctly identify serviceable posts.
Through our work, we believe technological interventions can help increase the interactions between police and citizens and thereby increase the trust people have on police. The police too may have a more directed and cost-labour efficient mechanism in dealing with any law and order situation reported on their Facebook page. This will increase the overall well-being and safety of society.
Full citation & link to the paper: Sachdeva, N., and Kumaraguru, P. Call for Service: Characterizing and Modeling Police Response to Serviceable Requests on Facebook. Accepted at the ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW), 2017. PDF