Algorithm 1: Bot detection via CB-MTE Framework
input: Twitter bot detection dataset T = {u1, u2, …, un}
output: Predicted labels $\mathrm{F}({\mathrm{x}}_{\mathrm{i}})\in$ (0: human, 1: bot)
1:	for each user ${\mathrm{u}}_{\mathrm{i}}$ in T do
2:	metadata feature extraction: ${\mathrm{M}}_{\mathrm{i}}={\mathrm{f}}_{\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{a}}({\mathrm{u}}_{\mathrm{i}})\in {ℝ}^{32}$
3:	textual feature extraction: ${\mathrm{f}}_{\mathrm{i}}\in {ℝ}^{768}$ ← Equations (1)–(4)
4:	Compute structural features for user ${\mathrm{u}}_{\mathrm{i}}$:
5:	$\mathrm{C}({\mathrm{u}}_{\mathrm{i}}),\mathrm{C}\mathrm{C}({\mathrm{u}}_{\mathrm{i}}),\mathrm{e}({\mathrm{u}}_{\mathrm{i}})$ ← Equations (5)–(7)
6:	graph feature extraction:
7:	${\mathrm{z}}_{\mathrm{i}}=\mathrm{D}\mathrm{e}\mathrm{e}\mathrm{p}\mathrm{W}\mathrm{a}\mathrm{l}\mathrm{k}({\mathrm{u}}_{\mathrm{i}})\in {ℝ}^{128}$ ← Equation (8)
8:	Concatenate with DeepWalk embedding:
9:	${\mathrm{h}}_{\mathrm{i}}$ ← Equation (9)
10:	Reduce dimensionality via UMAP:
11:	${\mathrm{f}}_{\mathrm{i}}^{\prime }\in {ℝ}^{16},{\mathrm{h}}_{\mathrm{i}}^{\prime }\in {ℝ}^{19}$ ← Equations (10)–(14)
12:	fused via vector concatenation: ${\psi }_{\mathrm{i}}=\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{c}\mathrm{a}\mathrm{t}\{{\mathrm{M}}_{\mathrm{i}},{\mathrm{f}}_{\mathrm{i}}^{\prime },{\mathrm{h}}_{\mathrm{i}}^{\prime }\}\in {ℝ}^{\mathrm{N}\times 67}$
13:	predict label using CatBoost classifier: $\mathrm{F}({\mathrm{x}}_{\mathrm{i}})$ ← Equations (15)–(22)
14:	end for