Social Network | The Appeal

Case study on how data science and specifically social networks can be used in a legal setting.

Introduction

Social Networks can serve a useful role in complex legal mattes such as the cause surrounding Murdock(murder and fraud) or the bankruptcy of Tom Girardi. These legal findings involve a myriad of stakeholders with varying interests. Moreover the justice system needs to exercise caution in understanding the requirements of these stakeholders to ensure that they are adequately compensated if/when a ruling is made in their favour. This is particularly the cause where means of compensation are limited such as in the case of bankruptcy.

Subsequently, using a fictional case study, this project will assess how a network could be used. This study will be based upon the novel, The Appeal. For those unaware of this book, its blurb is as follows:

“In a town full of secrets... Someone was murdered. Someone went to prison. And everyone's a suspect. Can you uncover the truth? Dear Reader - enclosed are all the documents you need to solve a case. It starts with the arrival of two mysterious newcomers to the small town of Lockwood, and ends with a tragic death. Someone has already been convicted of this brutal murder and is currently in prison, but we suspect they are innocent. What's more, we believe far darker secrets have yet to be revealed. Throughout the Fairway Players' staging of All My Sons and the charity appeal for little Poppy Reswick's life-saving medical treatment, the murderer hid in plain sight. Yet we believe they gave themselves away. In writing. The evidence is all here, between the lines, waiting to be discovered. Will you accept the challenge? Can you uncover the truth? The standout debut thriller of 2021 that delivers multiple brilliant twists, and will change the way you think about the modern crime novel.”(Waterstones, 2022)

It is told via message exchanges and as such shares similarities with evidence an investigator may face in real life.

Author’s notes: For ease, some of the information required for this project were collected in a google sheet. Others will be coded manually.

Overview

There are several networks which can be used to highlight the relations of those involved in this case. This includes those which highlight the relationship between parties as well as the number of correspondence directed towards one to another. In real life, these networks can hold information about how much money is owed to a party and the nature of the interaction such as victims, parties of interest or family.

In addition, there are also parties involved in a case with no/limited interest in the case. For example, the legal aids assessing the case to law enforcers who arrested, investigated and charged the suspect. In the case of The Appeal, these include the 3 individuals, Roderick Tanner, Charlotte Holroyd, Olufemi Hassan. Their network is as follows.

These individuals as well as organisations can be excluded from the main networks. The removal organisation relates to the issue that including it distorts the network. Theoretically, an email sent to a group email would have been received by those in the group. As such, depending on the contents should be either removed or assigned to those who would have received the information as a result of such a correspondence.

Relationship network

The first network which can illustrate the connection between parties in the case, is one which highlights the relationships: family, friends, colleagues. These are symmetric relationships for the purpose of this project but it may be a directed multigraph where relationships are more complex. For example, the relationship between Isabel and Samantha could be depicted as a friend but also as a stalker.

Based on this network the legal researcher can investigate the relationships between two individuals. For example, Samantha and Dr Szkatulska are colleagues.

In[8]: 
Bonds.get_edge_data("Samantha Greenwood","Dr Alicja Szkatulska",)["relationship"]
Out[8]:
'Colleague'

This network has a clustering of 0.8 and for Samantha in specific, based on her relationships, her local clustering coefficient is 0.52.

In [9]:nx.average_clustering(Bonds)
Out[9]:0.8119769593678511
In [10]:nx.clustering(Bonds, "Samantha Greenwood")
Out[10]:0.5228758169934641

For reference, transitivity is slightly low at 0.76 as individuals with a high number of connections such as Martin Hayward impacts the score.

In [11]:nx.transitivity(Bonds)
Out[11]:0.7603414313854235

Network and organisations

The case also highlights a plethora of groups from those who work at St Ann’s Hospital to those involved in the Fairway Players. Some individuals can be involved in different groups. In real life, outlining such networks can be used to assess if there is a conflict of interest. For example, in the Girardi bankruptcy, as Tom has been involved in law for a long time, it is vital to ensure that parties do not have a conflict. This was best highlighted by Erika Girardi’s attorney filing objections against Ronald Richards for his role in the bankruptcy proceedings. Using a bipartite graph, the relationship can be utilised.

In[21]: bipartite.is_bipartite(individuals)
Out[21]:True

This graph can show how via shared links to organisations including their past interactions with the overseas volunteer group, Samantha Greenwood and Dr Tish Bhatoa are connected, with other individuals. Weighting the graph can also highlight the strength of these connections.

This bipartite network also stores status of individuals such as whether a person is a suspect, victim etc. This once again may be used by the legal team where key information about an individual is needed to be stored and recalled effectively. For example, in the indictment of Jen Shah, with 10 other individuals also arrested, adding information to each node may aid in later reviews of information.

In[27]: individuals.nodes["Martin Hayward"]["status"]
Out[27]:'suspect'

Examination of those on the outer periphery of the network may also yield useful information for the investigation and its approach. Those on the boundary are more likely to have limited information about the case and may be more willing to share it due to their weak connection to the key individuals. In the case of this case study and the bipartite graph, these would include the following individuals.

In[31]:nx.periphery(individuals)# could be dropped from investigation
Out[31]:['Ravi Bhatoa', 'Ziggy Benjamin', 'Una', 'The Little Princess Trust']

In contrast, approaching those in the centre of the network should be done with caution, both in their depth of knowledge about the incident as well as in their links to suspects/ be the suspect. Whilst the network in a legal setting should already have knowledge of key individuals, it can also reveal additional information. For example, in the bipartite network, it has identified that some of the central figures/groups are as follows:

In[32]:nx.center(individuals)
Out[32]:['Martin Hayward',  'Helen Grace-Hayward',  'James Hayward',  'Sarah-Jane MacDonald',  'Kel Greenwood',  'Samantha Greenwood',  "John O'Dea",  'Jackie Marsh']

This list includes not only the victim, Samantha Greenwood but also individuals such as Kel Greenwood and Sarah-Jane MacDonald. These individuals are likely to be suspected of the crime as well as hold key information. In a real life example, the hope is that such a coding would reveal possible staff at Girardi law firm who may have been aware of the poor practices that were taking place.

Communication

Examining the actual evidence provided in the case study, there are two forms of networks that could be constructed.

Connection between the parties
Communication between parties

Connection between parties

If an email or message has been sent to and from two individuals, there is a connection between them. In addition, looking at the collections of messages logged, some responses to such correspondences from the Greenwoods are missing. Therefore this network simply logs if a message has been sent between the parties, irrespective of who sent the message. This creates the following graph.

Unpacking the stored data, reveals that between Martin Hayward and Dr Tish Bhatoa, there are 35 reported communications between them.

In[42]: G.get_edge_data('Martin Hayward', "Dr Tish Bhatoa")
Out[42]: {'count': 35}

In addition, whilst it is implied in the documents that Daniel Bhatoa and Samantha Greenwood share a relationship, in this network the shortest path between them involves Dr Tish Bhatoa and therefore has a length of 2.

In[43]: nx.shortest_path(G, "Samantha Greenwood", "Daniel Bhatoa")
Out[43]: ['Samantha Greenwood', 'Dr Tish Bhatoa', 'Daniel Bhatoa']

Aspects such as these, and understanding how possible connections can be formed in a network can provide law enforcement with new routes of investigation. For example, if the link between Samantha and Daniel was being investigated, questioning Trish, may provide key information. That said, in this case caution should be exercised as Tish and Daniel are related.

Focusing on Samantha, the paths connecting her and the other individuals can be described as follows using the Breadth-First search. Once again, aspects such as these may provided a point from which investigations and questioning could take place.

Communication between parties

As stated in the connection network, it glosses over the fact that these messages are directed towards an individual and does not separate where a reply was issued. A directed network on the other hand, captures this dynamic, as follows:

This graph enables the user to investigate how much a specific individual communicated to another party. For instance, of the 35 instances of communication previously noted, 16 were from Martin to Dr Bhatoa and 19 were from her to Martin.

In[52]: correspondance.get_edge_data('Martin Hayward', "Dr Tish Bhatoa")
Out[52]:{'Number of messages sent': 16.0, 'width': 1}
In [53]:correspondance.get_edge_data("Dr Tish Bhatoa",'Martin Hayward', )
Out[53]:{'Number of messages sent': 19.0, 'width': 1}

In this network, as a correspondences between third parties (ie those not directly linked with the key individuals are also included, it is not a linked to other parties in the network.) Understanding these aspects can identify possible parties that can be excluded to the investigation.

Relationship & Communication

Finally the relationship network can also be combined with the non-directed communication graph to highlight not only the relationship between parties but the frequency of communication. This network has been altered by removing the two individuals who appear to be unconnected to the majority of individuals in the graph.

In this newly formed network, the shortest path between parties is just under 3. The maximum distance is 5 and the distance is 3.

The list of eccentricities can be viewed as follows and highlights that there are a plethora of individuals who are on the outskirts of the network as per before.

In this more developed network, Samantha Greenwood, Martin Hayward, Isabel Beck are at the centre of the network. As Samantha is the victim, Martin and Isabel would be the key suspects who, if not responsible for the crime, are likely to hold key information about it.

In[71]: nx.center(G_updated)
Out[71]:
['Samantha Greenwood', 'Martin Hayward', 'Isabel Beck']

The graph has a connectivity of 1. Therefore aspects of the network appear to be weak. This weak bond may indicate that individuals may be more willing to disclose information in questioning as they do not feel indebted to one another.

In[73]:nx.edge_connectivity(G_updated)
Out[73]:1

The centrality of some of the key individuals in this network are as follows:

In[76]:degCent = nx.degree_centrality(G_updated)
       degCent["Martin Hayward"]
Out[76]:0.3875
In [77]:degCent["Sarah-Jane MacDonald"]
Out[77]:0.2625
In [78]:degCent["Samantha Greenwood"]
Out[78]:0.28750000000000003
In [79]:degCent["James Hayward"]
Out[79]:0.1125
In [80]:degCent["Isabel Beck"]
Out[80]:0.1875

They are all below the 0.5 level and are likely to be due to the number of groups involved and the manner in which different individuals are prominent in different groups. That said, based on closeness centrality, Martin Hayward is a prominent figure with a score of 0.59. In fact the only individual in the sample which had a closeness centrality below 0.5 was James. Therefore these figures can be considered prominent individuals.

In [81]:CloseCent = nx.closeness_centrality(G_updated)
In [82]:CloseCent["Martin Hayward"]
Out[82]:0.5882352941176471
In [83]:CloseCent["Sarah-Jane MacDonald"]
Out[83]:0.5095541401273885
In [84]:CloseCent["Samantha Greenwood"]
Out[84]:0.5263157894736842
In [85]:CloseCent["James Hayward"]
Out[85]:0.4419889502762431
In [86]:CloseCent["Isabel Beck"]
Out[86]:0.5095541401273885

For those who prefer to data visually, interactive graphs can be created. This graph can also highlight aspects such as the communities present in the network. These visualisations can further highlight how groups of individuals interact with one another.

Conclusion

This has been a study examining how social networks could be used in the legal system. As the examples used in this case indicated, networks are likely to serve a useful role in complex matters involving a plethora of parties, affecting a number of victims and/or covering several legal areas. Akin to the project, different networks can be utilised to service the needs of the legal services.

Furthermore analysis of such networks can also yield additional information concerning the interactions between parties and lines of possible investigation.

Additional Notes

This is the first time I have attempted a project such as this, not only using networks but also how it may be utilised in a legal setting. For the mode to be particularly versatile, improvements are required in both constructing the network as well as ensuring that extracting key information is done in a user-friendly manner. As it stands, a similar model will require an individual with knowledge of coding, which is a major limitation in the real life.