1: \begin{abstract}

2: This paper addresses the semantics of weighted argumentation graphs that are bipolar, i.e.\ 

3: contain both attacks and supports for arguments. It builds on previous work by 

4: Amgoud, Ben-Naim et. al.

5: \cite{DBLP:conf/kr/AmgoudB16,DBLP:conf/ijcai/AmgoudB16,DBLP:conf/ecsqaru/AmgoudB17,DBLP:conf/ijcai/AmgoudBDV17}. We study the various characteristics of acceptability semantics that

6: have been introduced in these works, and  introduce the notion 

7: of a \emph{modular acceptability semantics}.  A semantics is 

8:  modular if it cleanly separates \emph{aggregation} of

9: attacking and supporting arguments (for a given argument $a$) from the

10: computation of their \emph{influence} on $a$'s initial weight. 

11: We show that the

12: various semantics for bipolar argumentation graphs from the literature 

13: may be analysed as a composition of an aggregation function with an 

14: influence function. 

15: Based on this modular framework, we prove

16: general convergence and divergence theorems. We demonstrate that all well-behaved

17: modular acceptability semantics converge for all acyclic graphs and that no

18: \emph{sum}-based semantics can converge for all graphs. In particular,

19: we show divergence of Euler-based

20: semantics \cite{DBLP:conf/ecsqaru/AmgoudB17} for certain cyclic

21: graphs. Further,  we provide the first semantics for bipolar weighted

22: graphs that converges for all graphs.

23: \end{abstract}

24: