Omni-word Feature and Soft Constraint for Chinese Relation Extraction
Chen, Yanping and Zheng, Qinghua and Zhang, Wei

Article Structure

Abstract

Chinese is an ancient hieroglyphic.

Introduction

Information Extraction (IE) aims at extracting syntactic or semantic units with concrete concepts or linguistic functions (Grishman, 2012; McCallum, 2005).

Related Work

There are two paradigms extracting the relationship between two entities: the Open Relation Extraction (ORE) and the Traditional Relation Extraction (TRE) (Banko et al., 2008).

Feature Construction

In this section, the employed candidate features are discussed.

Discussion

In this section, we analyze the influences of employed feature sets and constraint conditions on the performances.

Conclusion

In this paper, We proposed a novel Omni-word feature taking advantages of Chinese sub-phrases.

Topics

entity mention

Appears in 21 sentences as: entity mention (15) entity mentions (8)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. An entity mention is a reference to an entity.
    Page 3, “Feature Construction”
  2. The entity mention is annotated with its full extent and its head, referred to as the extend mention and the head mention respectively.
    Page 3, “Feature Construction”
  3. Head Noun: The head noun (or head mention) of entity mention is manually annotated.
    Page 3, “Feature Construction”
  4. Position Feature: The position structure between two entity mentions (extend mentions).
    Page 3, “Feature Construction”
  5. Because the entity mentions can be nested, two entity mentions may have four coarse structures: “E1 is before E2”, “E1 is after E2”, “E1 nests in E2” and “E2 nests in El”, encoded as: ‘E1_B_E2’, ‘E1_A_E2’, ‘E1_N_E2’ and ‘E2_N_El’.
    Page 3, “Feature Construction”
  6. POS Tag: In our model, we use only the adjacent entity POS tags, which lie in two sides of the entity mention .
    Page 3, “Feature Construction”
  7. the POS tag with the adjacent entity mention information.
    Page 3, “Feature Construction”
  8. Position Sensitive: A position sensitive feature has a label indicating which entity mention it depends on.
    Page 4, “Feature Construction”
  9. IHS’ depend on the first entity mention .
    Page 4, “Feature Construction”
  10. To use the Omni-word feature, we segment each relation mention by two entity mentions .
    Page 5, “Feature Construction”
  11. Together with the two entity mentions, we get five parts: “FIRST”, “MIDDLE”, “END”, “E1” and “E2” (or less, if the two entity mentions are nested).
    Page 5, “Feature Construction”

See all papers in Proc. ACL 2014 that mention entity mention.

See all papers in Proc. ACL that mention entity mention.

Back to top.

relation extraction

Appears in 21 sentences as: (1) Relation Extraction (2) relation extraction (19)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. In this paper, we propose an Omni—word feature and a soft constraint method for Chinese relation extraction .
    Page 1, “Abstract”
  2. The results show a significant improvement in Chinese relation extraction , outperforming other methods in F-score by 10% in 6 relation types and 15% in 18 relation subtypes.
    Page 1, “Abstract”
  3. The performance of relation extraction is still unsatisfactory with a F-score of 67.5% for English (23 subtypes) (Zhou et al., 2010).
    Page 1, “Introduction”
  4. Chinese relation extraction also faces a weak performance having F-score about 66.6% in 18 subtypes (Dandan et al., 2012).
    Page 1, “Introduction”
  5. Therefore, the Chinese relation extraction is more difficult.
    Page 1, “Introduction”
  6. According to our survey, compared to the same work in English, the Chinese relation extraction researches make less significant progress.
    Page 1, “Introduction”
  7. Based on the characteristics of Chinese, in this paper, an Omni-word feature and a soft constraint method are proposed for Chinese relation extraction .
    Page 1, “Introduction”
  8. Propose a novel Omni-word feature for Chinese relation extraction .
    Page 1, “Introduction”
  9. There are two paradigms extracting the relationship between two entities: the Open Relation Extraction (ORE) and the Traditional Relation Extraction (TRE) (Banko et al., 2008).
    Page 2, “Related Work”
  10. In the field of Chinese relation extraction , Liu et al.
    Page 2, “Related Work”
  11. (2008) experimented with different kernel methods and inferred that simply migrating from English kernel methods can result in a bad performance in Chinese relation extraction .
    Page 2, “Related Work”

See all papers in Proc. ACL 2014 that mention relation extraction.

See all papers in Proc. ACL that mention relation extraction.

Back to top.

soft constraint

Appears in 19 sentences as: Soft Constraint (2) soft constraint (14) soft constraints (2) “soft constraint” (1)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. In this paper, we propose an Omni—word feature and a soft constraint method for Chinese relation extraction.
    Page 1, “Abstract”
  2. In order to utilize the structure information of a relation instance, we discuss how soft constraint can be used to capture the local dependency.
    Page 1, “Abstract”
  3. Both Omni-word feature and soft constraint make a better use of sentence information and minimize the influences caused by Chinese word segmentation and parsing.
    Page 1, “Abstract”
  4. and Soft Constraint
    Page 1, “Introduction”
  5. Based on the characteristics of Chinese, in this paper, an Omni-word feature and a soft constraint method are proposed for Chinese relation extraction.
    Page 1, “Introduction”
  6. Aiming at the Chinese inattentive structure, we utilize the soft constraint to capture the local dependency in a relation instance.
    Page 1, “Introduction”
  7. The soft constraints , proposed in this paper, are combined features like these syntactic or semantic constraints, which will be discussed in Section 3.2.
    Page 2, “Related Work”
  8. In our research, we proposed an Omni-word feature and a soft constraint method.
    Page 2, “Related Work”
  9. The soft constraint is the
    Page 2, “Feature Construction”
  10. 1If without ambiguity, we also use the terminology of “soft constraint” denoting features generated by the employed constraint conditions.
    Page 3, “Feature Construction”
  11. 3.2 Soft Constraint
    Page 4, “Feature Construction”

See all papers in Proc. ACL 2014 that mention soft constraint.

See all papers in Proc. ACL that mention soft constraint.

Back to top.

relation instance

Appears in 12 sentences as: relation instance (7) relation instances (5)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. In order to utilize the structure information of a relation instance , we discuss how soft constraint can be used to capture the local dependency.
    Page 1, “Abstract”
  2. Aiming at the Chinese inattentive structure, we utilize the soft constraint to capture the local dependency in a relation instance .
    Page 1, “Introduction”
  3. is obtained by segmenting every relation instance using the ICTCLAS package, collecting very word produced by ICTCLAS.
    Page 4, “Feature Construction”
  4. The structure information (or dependent information) of relation instance is critical for recognition.
    Page 4, “Feature Construction”
  5. Any relation instance violating these constraints (or below a predefined threshold) will be abandoned.
    Page 4, “Feature Construction”
  6. Deleting of relation instances is acceptable for open relation extraction because it always deals with a big data set.
    Page 4, “Feature Construction”
  7. For each of the remained sentences, we iteratively extract every entity mention pair as the arguments of relation instances for predicting.
    Page 5, “Feature Construction”
  8. After discarding the entity mention pairs that were used as positive instances, we generated 93,283 negative relation instances labelled as “OTHER”.
    Page 5, “Feature Construction”
  9. A maximum entropy multi-class classifier is trained and tested on the generated relation instances .
    Page 5, “Feature Construction”
  10. Second, most of relation instances have limited context.
    Page 6, “Feature Construction”
  11. They can precisely segment the relation instance into corresponding bins.
    Page 6, “Feature Construction”

See all papers in Proc. ACL 2014 that mention relation instance.

See all papers in Proc. ACL that mention relation instance.

Back to top.

POS tag

Appears in 12 sentences as: POS Tag (3) POS tag (6) POS tags (4)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. All the employed features are simply classified into five categories: Entity Type and Subtype, Head Noun, Position Feature, POS Tag and Omni-word Feature.
    Page 3, “Feature Construction”
  2. POS Tag: In our model, we use only the adjacent entity POS tags , which lie in two sides of the entity mention.
    Page 3, “Feature Construction”
  3. These POS tags are labelled by the ICTCLAS packagez.
    Page 3, “Feature Construction”
  4. The POS tags are not used independently.
    Page 3, “Feature Construction”
  5. the POS tag with the adjacent entity mention information.
    Page 3, “Feature Construction”
  6. In our experiment, the Head noun and POS Tag are utilized as position sensitive features, which has been introduced in Section 3.1.
    Page 4, “Feature Construction”
  7. Both head noun and POS tag are position sensitive.
    Page 5, “Feature Construction”
  8. The POS tags are referred to as fpos.
    Page 5, “Feature Construction”
  9. Except in Row 8 and Row 11, when two head nouns of entity pair were combined as semantic pair and when POS tag were combined with the entity type, the performances are decreased.
    Page 7, “Discussion”
  10. Comparing the reference set (5) with the reference set (3), the Head noan and adjacent entity POS tag get a better performance when used as singletons.
    Page 7, “Discussion”
  11. In this paper, for a better demonstration of the constraint condition, we still use the Position Sensitive as the default setting to use the Head noan and the adjacent entity POS tag .
    Page 7, “Discussion”

See all papers in Proc. ACL 2014 that mention POS tag.

See all papers in Proc. ACL that mention POS tag.

Back to top.

feature sets

Appears in 10 sentences as: Feature Set (1) feature set (4) feature sets (5)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. 3.1 Candidate Feature Set
    Page 3, “Feature Construction”
  2. To sum up, among the five candidate feature sets , the position feature is used as a singleton feature.
    Page 5, “Feature Construction”
  3. In the following experiments, focusing on Chinese relation extraction, we will analyze the performance of candidate feature sets and study the influence of the constraint conditions.
    Page 5, “Feature Construction”
  4. Five candidate feature sets are employed to generate the combined features.
    Page 5, “Feature Construction”
  5. The Omni-word feature set is denoted by fow.
    Page 5, “Feature Construction”
  6. In this section, we analyze the influences of employed feature sets and constraint conditions on the performances.
    Page 6, “Discussion”
  7. Because features may interact mutually in an indirect way, even with the same feature set , different constraint conditions can have significant influences on the final performance.
    Page 6, “Discussion”
  8. In Section 3, we introduced five candidate feature sets .
    Page 6, “Discussion”
  9. Symbol ‘7” means that the corresponding candidate features in the referential feature set are substituted by the new constraint condition.
    Page 7, “Discussion”
  10. By-Segmentation denotes the traditional segmentation based feature set generated by a segmentation tool, collecting every output of relation mention.
    Page 7, “Discussion”

See all papers in Proc. ACL 2014 that mention feature sets.

See all papers in Proc. ACL that mention feature sets.

Back to top.

entity type

Appears in 9 sentences as: Entity Type (2) entity type (5) Entity types (1) entity types (2)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. head noun, entity type , subtype, CLASS, LDC-TYPE, etc.)
    Page 3, “Feature Construction”
  2. In our experiment, the entity type , subtype and the head noun are used.
    Page 3, “Feature Construction”
  3. All the employed features are simply classified into five categories: Entity Type and Subtype, Head Noun, Position Feature, POS Tag and Omni-word Feature.
    Page 3, “Feature Construction”
  4. Entity Type and Subtype: In ACE 2005 RDC Chinese corpus, there are 7 entity types (Person, Organization, GPE, Location, Facility, Weapon and Vehicle) and 44 subtypes (e.g.
    Page 3, “Feature Construction”
  5. Those are generated by combining two entity types or two entity subtypes into a semantic pair.
    Page 4, “Feature Construction”
  6. type) and the second is a “Location” ( entity type ).
    Page 5, “Feature Construction”
  7. Entity types and subtypes are employed as semantic pair.
    Page 5, “Feature Construction”
  8. The entity type and subtype, head noun, position feature are referred to as fthpg.
    Page 5, “Feature Construction”
  9. Except in Row 8 and Row 11, when two head nouns of entity pair were combined as semantic pair and when POS tag were combined with the entity type , the performances are decreased.
    Page 7, “Discussion”

See all papers in Proc. ACL 2014 that mention entity type.

See all papers in Proc. ACL that mention entity type.

Back to top.

F-score

Appears in 6 sentences as: (1) F-score (5)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. The results show a significant improvement in Chinese relation extraction, outperforming other methods in F-score by 10% in 6 relation types and 15% in 18 relation subtypes.
    Page 1, “Abstract”
  2. The performance of relation extraction is still unsatisfactory with a F-score of 67.5% for English (23 subtypes) (Zhou et al., 2010).
    Page 1, “Introduction”
  3. Chinese relation extraction also faces a weak performance having F-score about 66.6% in 18 subtypes (Dandan et al., 2012).
    Page 1, “Introduction”
  4. F-score is computed by
    Page 5, “Feature Construction”
  5. In Row 2, with only the .7-"0w feature, the F-score already reaches 77.74% in 6 types and 60.31% in 18 subtypes.
    Page 6, “Feature Construction”
  6. In Table 3, it is shown that our system outperforms other systems, in F-score , by 10% on 6 relation types and by 15% on 18 subtypes.
    Page 6, “Feature Construction”

See all papers in Proc. ACL 2014 that mention F-score.

See all papers in Proc. ACL that mention F-score.

Back to top.

Chinese word

Appears in 6 sentences as: Chinese word (5) Chinese words (1)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. Both Omni-word feature and soft constraint make a better use of sentence information and minimize the influences caused by Chinese word segmentation and parsing.
    Page 1, “Abstract”
  2. The difficulty of Chinese IE is that Chinese words are written next to each other without delimiter in between.
    Page 1, “Introduction”
  3. Lacking of orthographic word makes Chinese word segmentation difficult.
    Page 1, “Introduction”
  4. (2008; 2010) also pointed out that, due to the inaccuracy of Chinese word segmentation and parsing, the tree kernel based approach is inappropriate for Chinese relation extraction.
    Page 2, “Related Work”
  5. Furthermore, for a single Chinese word , occurrences of 4 characters are frequent.
    Page 4, “Feature Construction”
  6. First, the specificity of Chinese word-formation indicates that the subphrases of Chinese word (or phrase) are also informative.
    Page 6, “Feature Construction”

See all papers in Proc. ACL 2014 that mention Chinese word.

See all papers in Proc. ACL that mention Chinese word.

Back to top.

maximum entropy

Appears in 5 sentences as: Maximum entropy (1) maximum entropy (4)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. We apply these approaches in a maximum entropy based system to extract relations from the ACE 2005 corpus.
    Page 1, “Introduction”
  2. The TRE systems use techniques such as: Rules (Regulars, Patterns and Propositions) (Miller et al., 1998), Kernel method (Zhang et al., 2006b; Zelenko et al., 2003), Belief network (Roth and Yih, 2002), Linear programming (Roth and Yih, 2007), Maximum entropy (Kambhatla, 2004) or SVM (GuoDong et al., 2005).
    Page 2, “Related Work”
  3. A maximum entropy multi-class classifier is trained and tested on the generated relation instances.
    Page 5, “Feature Construction”
  4. To implement the maximum entropy model, the toolkit provided by Le (2004) is employed.
    Page 5, “Feature Construction”
  5. Par in Column 4 is the number of parameters in the trained maximum entropy model, which indicate the model complexity.
    Page 7, “Discussion”

See all papers in Proc. ACL 2014 that mention maximum entropy.

See all papers in Proc. ACL that mention maximum entropy.

Back to top.

word segmentation

Appears in 5 sentences as: word segmentation (5)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. The Omni-word feature uses every potential word in a sentence as lexicon feature, reducing errors caused by word segmentation .
    Page 1, “Abstract”
  2. Both Omni-word feature and soft constraint make a better use of sentence information and minimize the influences caused by Chinese word segmentation and parsing.
    Page 1, “Abstract”
  3. Lacking of orthographic word makes Chinese word segmentation difficult.
    Page 1, “Introduction”
  4. (2008; 2010) also pointed out that, due to the inaccuracy of Chinese word segmentation and parsing, the tree kernel based approach is inappropriate for Chinese relation extraction.
    Page 2, “Related Work”
  5. On the other hand, the Omni-word can avoid these problems and take advantages of Chinese characteristics (the word-formation and the ambiguity of word segmentation ).
    Page 4, “Feature Construction”

See all papers in Proc. ACL 2014 that mention word segmentation.

See all papers in Proc. ACL that mention word segmentation.

Back to top.

tree kernel

Appears in 5 sentences as: tree kernel (5)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. (2012) proposed a convolution tree kernel .
    Page 2, “Related Work”
  2. (2010) employed a model, combining both the feature based and the tree kernel based methods.
    Page 2, “Related Work”
  3. (2008; 2010) also pointed out that, due to the inaccuracy of Chinese word segmentation and parsing, the tree kernel based approach is inappropriate for Chinese relation extraction.
    Page 2, “Related Work”
  4. The reason of the tree kernel based approach not achieve the same level of accuracy as that from English may be that segmenting and parsing Chinese are more difficult and less accurate than processing English.
    Page 2, “Related Work”
  5. In this field, the tree kernel based method commonly uses the parse tree to capture the structure information (Zelenko et al., 2003; Culotta and Sorensen, 2004).
    Page 4, “Feature Construction”

See all papers in Proc. ACL 2014 that mention tree kernel.

See all papers in Proc. ACL that mention tree kernel.

Back to top.

n-Gram

Appears in 5 sentences as: N-Gram (1) n-Gram (4)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. Despite the Omni-word can be seen as a subset of n-Gram feature.
    Page 4, “Feature Construction”
  2. It is not the same as the n-Gram feature.
    Page 4, “Feature Construction”
  3. N-Gram features are more fragmented.
    Page 4, “Feature Construction”
  4. In most of the instances, the n-Gram features have no semantic meanings attached to them, thus have varied distributions.
    Page 4, “Feature Construction”
  5. Because Chinese has plenty of characters5, when the corpus becoming larger, the n-Gram (ng,4) method is difficult to be adopted.
    Page 4, “Feature Construction”

See all papers in Proc. ACL 2014 that mention n-Gram.

See all papers in Proc. ACL that mention n-Gram.

Back to top.

feature space

Appears in 5 sentences as: feature space (5)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. (2010) proposed a model handling the high dimensional feature space .
    Page 2, “Related Work”
  2. Because the number of lexicon entry determines the dimension of the feature space , performance of Omni-word feature is influenced by the lexicon being employed.
    Page 3, “Feature Construction”
  3. Combining two head nouns may increase the feature space
    Page 7, “Discussion”
  4. Such a large feature space makes the occurrence of features close to a random distribution, leading to a worse data sparseness.
    Page 7, “Discussion”
  5. The size of the employed lexicon determines the dimension of the feature space .
    Page 8, “Conclusion”

See all papers in Proc. ACL 2014 that mention feature space.

See all papers in Proc. ACL that mention feature space.

Back to top.

manually annotated

Appears in 3 sentences as: manually annotated (3)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. Disadvantages of the TRE systems are that the manually annotated corpus is required, which is time-consuming and costly in human labor.
    Page 2, “Related Work”
  2. Head Noun: The head noun (or head mention) of entity mention is manually annotated .
    Page 3, “Feature Construction”
  3. Third, the entity mentions are manually annotated .
    Page 6, “Feature Construction”

See all papers in Proc. ACL 2014 that mention manually annotated.

See all papers in Proc. ACL that mention manually annotated.

Back to top.

SVM

Appears in 3 sentences as: SVM (3)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. The TRE systems use techniques such as: Rules (Regulars, Patterns and Propositions) (Miller et al., 1998), Kernel method (Zhang et al., 2006b; Zelenko et al., 2003), Belief network (Roth and Yih, 2002), Linear programming (Roth and Yih, 2007), Maximum entropy (Kambhatla, 2004) or SVM (GuoDong et al., 2005).
    Page 2, “Related Work”
  2. (2005) introduced a feature based method, which utilized lexicon information around entities and was evaluated on Winnow and SVM classifiers.
    Page 2, “Related Work”
  3. For each type of these relations, a SVM was trained and tested independently.
    Page 2, “Related Work”

See all papers in Proc. ACL 2014 that mention SVM.

See all papers in Proc. ACL that mention SVM.

Back to top.

Chinese word segmentation

Appears in 3 sentences as: Chinese word segmentation (3)
In Omni-word Feature and Soft Constraint for Chinese Relation Extraction
  1. Both Omni-word feature and soft constraint make a better use of sentence information and minimize the influences caused by Chinese word segmentation and parsing.
    Page 1, “Abstract”
  2. Lacking of orthographic word makes Chinese word segmentation difficult.
    Page 1, “Introduction”
  3. (2008; 2010) also pointed out that, due to the inaccuracy of Chinese word segmentation and parsing, the tree kernel based approach is inappropriate for Chinese relation extraction.
    Page 2, “Related Work”

See all papers in Proc. ACL 2014 that mention Chinese word segmentation.

See all papers in Proc. ACL that mention Chinese word segmentation.

Back to top.

PDF available from www.aclweb.org.
Send us an email