Data Science Seminar on CPDP Database

Team: the dapper squirrels

Table of Contents

• Theme
• Relational Analysis Questions
• Verification Technique
  • Visualization
  • Interactive Visualization
  • Graph Analytics
  • NLP models
• References
• License

Theme

Misconduct analysis in terms of different locations and communities can be valuable. Is there over-policing in low socio-eco status neighborhoods? We could compare the low-income area data with high in-come area data. The income of the neighbor could be a factor to influence the “victim” narrative (complaint report). We plan to dive deep into the relationship between location, income level, and police misconduct.

Our final report: Final Report;
Our final presentation: Final Presentation.

Relational Analysis Questions

1. What are the TOP5 richest and lowest income neighborhoods?
2. What are the neighborhoods’ income and CRs(complaint record) per capita?
3. What is the TRRS(tactical response report) per capita?
4. What is the percentage of each race in the community?
5. What are the top 5 streets in allegation counts for each beat area?

Please check out SQLs and findings in checkpoint-1.

Verification Technique

Visualization

We would like to dive deep into the relationship between the income of officers versus common people by visualization tools like Tableau, which is more clear and interpretable than numbers.

1. Visualize by using a line chart to show the officer hours/year per capita in each community by years if change over time.
2. Scatterplot of Complaint Report per capita V.S. Tactical Respond Report per capita. We could also consider lawsuits between “victims” and police officers; search warrants granted in each complaint?

Please check out our code and findings in checkpoint-2.

Interactive Visualization

Looking into trends of comparisons and numbers in time can be intuitive. We are specifically interested in the comparisons of complaint rate change as opposed to income change in time.

1. Highlighting the high and low socio-economy status communities with different colors and plot TRRs on them. Set up a time slider to see how it changes over time.
2. Using color code(heat map) of A&A (dara_officer assignment attendance) in different neighborhoods. Set up a time slider to see how it changes over time.

Please check out our code and findings in checkpoint-3.

Graph Analytics

Graph analytics can be very useful in analyzing relationships between different groups of people. We can create nodes based on their income, race, neighborhood, and other attributes. After building the graph, we can analyze interactions among different nodes and even graphlets.

1. Making nodes of officers and victims by their income, race, locations, and even unsupervised machine learning models to learn the cluster and see if there is a potential connection between officers and victims.
2. Network dynamics of co-accused in each cohort can be interesting. The analytics can be done with the following:
   1. Make use of Triangle Count Algorithms for each cohort.
   2. Make use of the Page Rank Algorithm to find the most connected officer in all cohorts.
   3. How many CRs that officers have and how many co-accused for each cohort.
   4. Compare the top k largest cohort of police officers in high and low socio-economy status.

Please check out our code and findings in checkpoint-4.

NLP models

Topic modeling is quite popular and useful in the NLP area. We are interested in topics in CRS in each cohort. Just as the descriptions put, manual labeling by officers is fallacious in many cases. However, with the development of natural language processing, we may apply it to label complaint texts with high accuracy. This would help a lot in understanding and analyzing misconduct. Technical solutions span classical algorithms like TF-IDF and cutting-edge research methods like graph neural networks [2] and BERT [3]. We plan to dive deeper into the area and find out more effective and proper methods for us.

Please check out our code and findings in checkpoint-5.

References

[1] Liang et al. EURASIP Journal on Wireless Communications and Networking (2017) 2017:211

[2] O. Alqaryouti, T. A. Farouk, A. R. Nabhan and K. Shaalan, “Graph-Based Keyword Extraction,” in Intelligent Natural Language Processing: Trends and Applications, Springer, Cham, 2018, pp. 159-172. [DOI:10.1007/978-3-319-67056-0_9]

[3] Keyword Extraction with BERT https://towardsdatascience.com/keyword-extraction-with-bert-724efca412ea

License

This project is licensed under the MIT License - see the LICENSE file for details.