This is not a java3D specific answer.
In general a matrix can be built such that there are 4 vectors that describe it.
1) A side (or lateral) vector
2) An up vector
3) A direction vector
4) A position
Each row of a 4x4 matrix.
Thus for a simple identity matrix we have the following matrix (I'll define a column major matrix, for a row major matrix all you need to do is swap the matrix indices around such that row 2 col 3 becomes row 3 col 2 throughout the matrix).
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
in this the first column is the side vector. The second column the up vector. The third the direction and the fourth the position.
Logically we can see that the vector (1, 0, 0, 0) points along the x axis (and thus is the side vector). The vector (0, 1, 0, 0) points along the y axis (and thus is the up vector). The third (0, 0, 1, 0) points along the Z-axis (and thus is the direction vector). The fourth (0, 0, 0, 1) indicates that the objects does not move at all.
Now lets say we wanted to face along the X-axis.
Obviously that would mean we have a vector of (1, 0, 0, 0 ) for our direction vector. Up would still be (0, 1, 0, 0) and position still 0, 0, 0 1. So what would our side vector be? Well, logically it would point along the z-axis. But which way? Well hold your fingers such that one finger points forward, one to the side and one up. Now rotate so that the forward finger is facing the same direction as the side pointing finger. Which way is the side pointing finger pointing now? The opposite direction to the original direction pointing finger. Thus the matrix is
0 0 1 0
0 1 0 0
-1 0 0 0
0 0 0 1
At this point things seemingly get a little more complicated. It is simple enough to take an arbitrary position and an arbitrary point to look at (I'll call them vPos and vFocus). It is easy enough to form a vector from vPos to vFocus by subtracting vPos from vFocus (vFocus.x - vPos.x, vFocus.y - vPos.y, vFocus.z - vPos.z, vFocus.w - vPos.w ). Bear in mind all positions should be defined with a '1' in the w position where all directions should have a '0'. This is automatically taken care of when you do the subtraction above as the 1 in both ws will cancel out and leave 0. Anyway, we now have a vector pointing from the position towards vFocus we'll call it vDir. Unfortunately it has the length of the difference between vPos and vFocus. However if we divide the vDir vector by its length (vDir.x / length, vDir.y / length, vDir.z / length, vDir.w / length) then we normalise it and we have a direction with a total length of 1.
At this ponit we now have our 3rd and 4th columns of our matrix. Now, lets assuem up is still (0, 1, 0, 0) or vUp. We can assume that the crossproduct of the direction and vUp will produce a vector that is perpendicular (and also of unit length) to the plane formed by vDir and vUp. This gives us our side vector or vLat. Now .. we did kind of assume the up vector so its not strictly correct. We can now calculate it exactly by taking the cross product of vLat and vDir and we have all 4 vectors.
The final matrix is thus defined as follows
vLat.x vUp.x vDir.x vPos.x
vLat.y vUp.y vDir.y vPos.y
vLat.z vUp.z vDir.z vPos.z
vLat.w vUp.w vDir.w vPos.w
This isn't strictly the full answer as you will get problems as you look towards a point near to your (0, 1, 0, 0) vector but that should work for most cases.