GNN based unsupervised Anomaly Detection for Heterogenous Graphs

I am working on a project where I am doing Unsupervised Anomaly Detection on employee expenses on HCP transfer Of Value. I am trying to use Graph Neural Network to detect anomalies with proper explainability. So let me give you a context about the data and the type of risks we are trying detect. I will also provide some nodes and relation/ edges that comes out of my mind. Help me build proper edge node and edge and node features and relationship based on your real world knowledge on HCP Transfer Of Values in real world. We have a transaction attendee level data. Each employee does some meetings or meals with HCPs and upload these expenses on concur sytem. So employee submits a report which contains multiple entries/ transactions. Each transactions ar now divided into rows of attendee, so for each transaction lets say we have 3 HCPs, then for each transaction we will have three rows for each attendee attributing spends/ costs to each attendee. We also have columns of total spend of an entries. There are two types of transaction credit card and cash. All transactions have the following info- Employee, Spend, Type Of Expense, Transaction Date, Location , Home Country, Merchant/vendor of the transaction. We can also get the employee job profile info and for attendees the type of attendee. Main focus is to identify unusual Rep, HCP patterns. I need idea on relationships/ edge / node features. Right now I have thought of these nodes- employee/rep, report, transaction, merchant/vendor, attendee. This type of relationships I am thinking of Employee submits report. Report contains cash/ credit card transaction owing to vendor/merchant. Employee performs transaction having x amount located in this country, city. Transaction contains attendee. Attendee is related to employee through transaction, want build direct connection. Give me step by step proper idea on -

1. what more entities/ nodes we should consider?
2. What more relations we can think of?
3. What can be the node and edge features?
4. What models can be applied- unsupervised techniques for Anomaly Detection?
5. How to explain anomalies- what caused it anomaly? Build network, graph and models in such a way that it will have more advantage over traditional ML based unsupervised models. Please explain how graph methods can detect anomalies better than traditional ML models. Also if i want integrate sequential/ time component patterns or temporal patterns in graph how can i do that