We use the Millennium simulation to show that halo clustering varies significantly with cosmic web type .
Halos are classified as node , filament , sheet and void halos based on the eigenvalue decomposition of the velocity shear tensor .
The velocity field is sampled by the peculiar velocities of a fixed number of neighbouring halos and spatial derivatives are computed using a kernel borrowed from smoothed particle hydrodynamics .
The classification scheme is used to examine the clustering of halos as a function of web type for halos with masses larger than 10 ^ { 11 } .
We find that node halos show positive bias , filament halos show negligible bias , and void and sheet halos are anti-biased independent of halo mass .
Our findings suggest that the mass dependence of halo clustering is rooted in the composition of web types as a function of halo mass .
The substantial fraction of node type halos for halo masses \gtrsim 2 \times 10 ^ { 13 } h ^ { -1 } M _ { \odot } leads to positive bias .
Filament type halos prevail at intermediate masses , 10 ^ { 12 } -10 ^ { 13 } h ^ { -1 } M _ { \odot } , resulting in unbiased clustering .
The large contribution of sheet type halos at low halo masses \lesssim 10 ^ { 12 } h ^ { -1 } M _ { \odot } generates anti-biasing .