Yes, remove the bad items and prefs from your backing store. That gets updated then next time you refresh(). Eventually, you want those gone, it seems, since they're not useful information.
Among other things, the Rescorer is there to filter out items that are useful for the internal CF algorithm, but not a valid recommendation (e.g. out-of-stock items). But it could be slightly abused to filter out bad data on the fly, on the front end. You don't need to or want to attach additional info to those user-item associations. Presumably the bad-ness is a property of the item. So you just need a Rescorer that knows a list of bad item IDs, and filters them. On Mon, Oct 17, 2011 at 1:47 PM, Octavian Covalschi <[email protected]> wrote: > I understand. Even though I don't have feed files, we're using MongoDB as > the main data source, I can update database for the "old" or "out of stock" > items... > > So even in the case of ghost items or those I want to exclude and use > Rescorer, I still need to update the in memory model, right? As far I > understand it's not enough to update the feed, unless It's being > refreshed... Is there a way to store additional optional data along with > userId <-> itemId? The only solution I see now is to extend > GenericPreference and add additional properties like isOld, price (for other > purposes), is this a correct approach? > > PS: Thanks guys for your time! > > On Mon, Oct 17, 2011 at 2:35 AM, Sean Owen <[email protected]> wrote: > >> In general this DataModel is not modifiable, which is why you're >> having to hack at it. The idea is to update the underlying data, >> ideally -- for example, the file containing the data, perhaps by using >> update (delta) files. >> >> Your approach works though. There is not a better way to do it, if >> this is what you're doing. >> >> As Ted mentioned, keep data on items that are real and valid, just out >> of stock or old or something. You can filter them using a Rescorer so >> that they don't end up in the output. >> >> You can do the same thing for ghost data, if you like. That at least >> means you have an immediate way of filtering and can wait for a data >> reload or update rather than do it every time on the fly. >> >> On Mon, Oct 17, 2011 at 2:58 AM, Octavian Covalschi >> <[email protected]> wrote: >> > Hi there. >> > >> > We have the taste war running and we are trying to accomplish a realtime >> > functionality by removing/adding user items on the fly. So far adding and >> > removing items for particular users has been working fine for us and >> > recommendations are being changed on the fly.. though it's still in >> beta... >> > The last piece is to remove an item from ALL users, it's necessary when >> the >> > item is removed from the system... So far I've got this: >> > >> > public void removeItemFromAllUsers(long itemId) throws >> NoSuchItemException { >> > PreferenceArray prefArray = ((GenericDataModel) >> > delegate).getPreferencesForItem(itemId); >> > >> > if (prefArray.length() > 0) { >> > FastByIDMap<PreferenceArray> rawData = ((GenericDataModel) >> > delegate).getRawUserData(); >> > >> > for (Preference pr : prefArray) { >> > PreferenceArray prefs = rawData.get(pr.getUserID()); >> > >> > rawData.remove(pr.getUserID()); >> > >> > if (prefs.length() > 1) { >> > PreferenceArray newPrefs = new >> > GenericUserPreferenceArray(length - 1); >> > for (int i = 0, j = 0; i < length; i++, j++) { >> > if (prefs.getItemID(i) == itemId) { >> > j--; >> > } else { >> > newPrefs.set(j, prefs.get(i)); >> > } >> > } >> > rawData.put(pr.getUserID(), newPrefs); >> > } >> > } >> > delegate = new GenericDataModel(rawData); >> > } >> > } >> > >> > I'm still pretty new with mahout and was wondering if there is a better >> way, >> > in particular make use of getRawUserData(), which I tried but it didn't >> > help... >> > >> > Thank you in advance. >> > >> >
