:Configure re-ranking

ItemRankScore

ItemRankScore sorts all items in descending order by score. It is built into the DPI engine — no separate configuration is needed. Reference it directly in SortNames.

Boost score sort (BoostScoreSort)

Use BoostScoreSort to boost or demote item scores after fine-grained sorting, based on item or user properties. For example, to surface women's apparel higher for female users, multiply the score for items where sex = female. To suppress low-quality items, apply score * 0.5 when a quality flag is below a threshold.

The policy applies a conditional expression to adjust scores:

• Conditions: match items or users by property (for example, sex = female, category = electronics)

• Expression: a formula applied to the score when conditions are met (for example, score * 1.5, score * 0.5)

Configuration example:

{
    "SortConfs": [
        {
            "Name": "BoostScoreSort",
            "SortType": "BoostScoreSort",
            "Debug": false,
            "BoostScoreConditions": [
                {
                    "Conditions": [
                        {
                            "Name": "sex",
                            "Domain": "item",
                            "Type": "string",
                            "Value": "gender",
                            "Operator": "equal"
                        }
                    ],
                    "Expression": "score * 2"
                }
            ]
        }
    ]
}

This configuration multiplies the score by 2 for items where the sex feature equals male.

BoostScoreSort fields

FilterParamConfig fields

For more information about conditional settings, see Adjust count filter (AdjustCountFilter).

Boost score by weight (BoostScoreByWeight)

Use BoostScoreByWeight when different items need different score multipliers stored in a Hologres table, rather than a rule-based expression. The score formula is weight × item.score.

Configuration example:

{
    "SortConfs": [
        {
            "Name": "BoostScoreByWeight",
            "SortType": "BoostScoreByWeight",
            "TimeInterval": 172800,
            "BoostScoreByWeightDao": {
                "AdapterType": "hologres",
                "HologresName": "pai_rec",
                "HologresTableName": "test",
                "ItemFieldName": "item_id",
                "WeightFieldName": "weight"
            }
        }
    ]
}

BoostScoreByWeightDao fields

Diversity rule sort (DiversityRuleSort)

Use DiversityRuleSort to prevent recommendations from clustering around a single category, author, or tag. The policy enforces spacing rules across one or more item properties.

DiversityRuleSort must be used with the ExcludeRecalls parameter.

Key concepts

• Diversification dimension: The item property to diversify on, such as category, author, or tag.

• IntervalSize (k): The maximum number of consecutive items from the same dimension value. For example, IntervalSize: 2 means no more than 2 consecutive items can share the same category.

• WindowSize (n) and FrequencySize (m): Within a sliding window of n items, an item from the same dimension value cannot appear more than m times. For example, WindowSize: 10, FrequencySize: 2 means the same category can appear at most twice in any 10 consecutive positions.

Diversity rules apply to a single request only — cross-request diversity is not enforced.

Configuration example:

{
    "SortConfs": [
        {
            "Name": "DiversityRuleSort",
            "SortType": "DiversityRuleSort",
            "DiversitySize": 100,
            "DiversityRules": [
                {
                    "Dimensions": ["spfl"],
                    "WindowSize": 10,
                    "FrequencySize": 1
                }
            ],
            "ExcludeRecalls": [
                "ColdStartVideoVectorRecall",
                "LinUcbRecall_default2"
            ],
            "Conditions": [
                {
                    "Name": "spflPick",
                    "Domain": "user",
                    "Type": "string",
                    "Value": "",
                    "Operator": "equal"
                }
            ]
        }
    ]
}

DiversityRuleSort fields

Selection logic

By default, an item is output only if it satisfies all diversity rules. If no candidate satisfies all rules, the first candidate found is selected.

With weighted rules, if no candidate satisfies all rules, the candidate satisfying the highest-weighted rules is selected. Ties are broken by position in the candidate list.

Example: exclusion rules and search depth

Items with tag = t1 are excluded from positions 1–4. Positions 1–4 still respect diversity rules, but items with tag = t1 cannot fill those positions.

{
    "Name": "DiversityRuleSort",
    "SortType": "DiversityRuleSort",
    "DiversityRules": [
        {
            "Dimensions": ["tag"],
            "WindowSize": 5,
            "FrequencySize": 1
        }
    ],
    "ExclusionRules": [
        {
            "Positions": [1, 2, 3, 4],
            "Conditions": [
                {
                    "Name": "tag",
                    "Domain": "item",
                    "Type": "string",
                    "Value": "t1",
                    "Operator": "equal"
                }
            ]
        }
    ],
    "ExploreItemSize": 200
}

Example: weighted diversity rules

{
    "Name": "DiversityRuleSort",
    "SortType": "DiversityRuleSort",
    "DiversityRules": [
        {
            "Dimensions": ["tag"],
            "WindowSize": 5,
            "FrequencySize": 1,
            "Weight": 1
        },
        {
            "Dimensions": ["category"],
            "WindowSize": 3,
            "FrequencySize": 1,
            "Weight": 3
        }
    ],
    "ExclusionRules": [
        {
            "Positions": [1],
            "Conditions": [
                {
                    "Name": "tag",
                    "Domain": "item",
                    "Type": "string",
                    "Value": "t1",
                    "Operator": "equal"
                }
            ]
        }
    ]
}

DPPSort

DPPSort applies the Determinantal Point Process (DPP) algorithm to balance relevance and diversity. For background on the algorithm, see An intuitive understanding of the DPP-based algorithm for improving recommendation diversity.

Prerequisites

DPPSort requires item embedding vectors that represent item content — not behavioral similarity.

• Use: image embeddings, text description embeddings, or embeddings from static item attributes (categories, properties)

• Avoid: embeddings trained from user behavioral data

The dimensions you want to diversify must be captured in the embeddings. For example, to diversify on price, the price feature must be included when training the embedding model.

Configuration example:

{
    "SortConfs": [
        {
            "Name": "DPPSort",
            "SortType": "DPPSort",
            "DPPConf": {
                "Name": "DPPSort",
                "DaoConf": {
                    "AdapterType": "hologres",
                    "HologresName": "geeko_rec"
                },
                "TableName": "item_embedding_metric_learning",
                "TableSuffixParam": "embedding_date",
                "TablePKey": "product_id",
                "EmbeddingColumn": "embedding",
                "Alpha": 4.5,
                "NormalizeEmb": "false",
                "WindowSize": 10
            }
        }
    ]
}

DPPConf fields

SSDSort

SSDSort applies the Structured Self-Distillation (SSD) algorithm as an alternative to DPP. For background, see Improving the diversity of recommendation results: an analysis of MMR/DPP/SSD principles.

SSDSort has the same prerequisites as DPPSort — item embedding vectors must represent content, not behavioral similarity. The key difference is the tuning parameter: Gamma (larger value = more diversity), compared to DPP's Alpha (larger value = more relevance).

Configuration example:

{
    "SortConfs": [
        {
            "Name": "SSDSort",
            "SortType": "SSDSort",
            "SSDConf": {
                "Name": "SSDSort",
                "DaoConf": {
                    "AdapterType": "hologres",
                    "HologresName": "geeko_rec"
                },
                "TableName": "item_embedding_metric_learning",
                "TablePKey": "item_id",
                "EmbeddingColumn": "embedding",
                "Gamma": 0.25,
                "UseSSDStar": true,
                "NormalizeEmb": "false",
                "MinScorePercent": 0.1,
                "CandidateCount": 200,
                "WindowSize": 5
            }
        }
    ]
}

SSDConf fields

Multi-channel recall sort (MultiRecallMixSort)

Use MultiRecallMixSort when you have multiple recall channels and need to control how their results are interleaved in the final output. Common use cases:

• Guarantee minimum exposure for cold-start items

• Pin items from a specific recall channel to fixed positions

Configuration example (recall-name-based mixing):

{
    "SortConfs": [
        {
            "Name": "MixSort",
            "SortType": "MultiRecallMixSort",
            "RemainItem": false,
            "MixSortRules": [
                {
                    "MixStrategy": "random_position",
                    "NumberRate": 0.1,
                    "RecallNames": ["OTSGlobalHot"]
                },
                {
                    "MixStrategy": "fix_position",
                    "Positions": [1, 3, 5],
                    "RecallNames": ["RecallName1"]
                }
            ]
        }
    ]
}

You can also select items by item property conditions instead of recall names:

{
    "SortConfs": [
        {
            "Name": "MixSortByItemFeature",
            "SortType": "MultiRecallMixSort",
            "RemainItem": false,
            "MixSortRules": [
                {
                    "MixStrategy": "random_position",
                    "NumberRate": 0.1,
                    "Conditions": [
                        {
                            "Name": "gender",
                            "Domain": "item",
                            "Type": "string",
                            "Value": "man",
                            "Operator": "equal"
                        }
                    ]
                }
            ]
        }
    ]
}

MultiRecallMixSort fields