1: \begin{abstract}

2: We develop a complete theory 

3: for the combinatorics of 

4: walk-counting on a directed graph 

5: in the case where each

6: backtracking step is downweighted

7: by a given factor.

8: By deriving expressions for the 

9: associated generating functions, we  

10: also obtain 

11: linear systems for computing 

12: centrality measures in this setting.

13: In particular,   

14:  we show that

15:  backtrack-downweighted 

16:   Katz-style network 

17:   centrality can be computed at the same cost 

18:    as standard Katz.

19:    Studying the limit of this 

20:     centrality measure at its radius of convergence 

21:     also leads to a new expression for 

22:      backtrack-downweighted  eigenvector 

23:      centrality that generalizes

24:      previous work to the case where directed edges are present.

25:         The new theory allows us to 

26:        combine advantages 

27:        of standard and nonbacktracking 

28:        cases, avoiding localization while accounting for tree-like 

29:         structures. 

30:      We illustrate the behaviour of the

31:        backtrack-downweighted centrality measure on both synthetic

32:       and real networks. 

33:  \end{abstract}

34: